Document [original]

Designing Liner Shipping Feeder Service

Networks in the New Era of Mega Containerships

vorgelegt von

M.Sc.

Olcay Polat

aus Denizli

von der Fakultät VII- Wirtschaft und Management

der Technischen Universität Berlin

zur Erlangung des akademischen Grades

Doktor der Ingenieurwissenschaften

- Dr.-Ing. -

genehmigte Dissertation

Promotionsausschuss:

Vorsitzender: Prof. Dr. F. Straube

Berichter: Prof. Dr. H.-O. Günther

Berichter: Prof. Dr. O. Kulak

Tag der wissenschaftlichen Aussprache: 26. Juni 2013

Berlin 2013

D 83

Eidesstattliche Erklärung

Hiermit erkläre ich an Eides statt, daß ich die vorliegende Dissertation selbständig

und nur unter Verwendung der angegebenen Hilfsmittel angefertigt habe. Die den be-

nutzten Quellen wörtlich oder inhaltlich entnommenen Stellen habe ich als solche

kenntlich gemacht.

Berlin, den 01.05.2013

Olcay Polat

iii

Acknowledgements

I would first of all like to express my deepest gratitude to my advisor, Prof. Dr. Hans-

Otto Günther for his continuous support, guidance and patience throughout my Ph.D.

study. This dissertation could not have been written without Prof. Dr. Hans-Otto Gün-

ther who is a dedicated and encouraging advisor. Also, I would like to thank my second

advisor, Prof. Dr. Osman Kulak who opened my way at Technical University of Berlin

and helped me throughout the progress of this dissertation.

Moreover, I would like to express my gratitude to Prof. Dr. Kap Hwan Kim, Prof.

Dr. Ceyda Oğuz, Prof. Dr. Marielle Christiansen and Assoc. Prof. Dr. Bülent Çatay for

their helpful comments and advice. My thanks also go to Rico Gujjula, Mario Lueb,

Andreas Schöpperl and Maren Kühl from the Department of Production Management at

Technical University of Berlin for encouragement and for making the average work day

more fun and interesting. In addition, I would like to say that I am grateful to all my

instructors and professors in Eskişehir Osmangazi University and Pamukkale University

for them equipping me with their knowledge and academic skills.

I also want to specially thank my friends, Senem Acar Kübart, Thore Kübart, Simge

Tokat, Derya Kuert, Vildan Sirkeci, Filiz Canlı, Mehmet Canbeyli, Mustafa Egemen

Taner, Ozan Çapraz, Bayram Kara and Can Berk Kalaycı for their friendship, encour-

agement and support in Berlin. Special thanks to my parents, İbrahim and Türkan Polat,

and my brother, Hüseyin Polat, for their love, support and understanding. Finally, I

would like to express my special gratitude to my fiancée, my true love, Leyla Özgür for

her love, encouragement, endless patience and support throughout all my Ph.D. years.

This work was partially supported by the German Academic Exchange Service

(DAAD) with the grant number A/08/77565 and Ministry of Transport, Maritime Af-

fairs and Communications of the Republic of Turkey.

May 2013, Berlin

Olcay Polat

Table of Contents

Eidesstattliche Erklärung .................................................................................................. ii

Acknowledgements .......................................................................................................... iii

Table of Contents ............................................................................................................. iv

List of Figures .................................................................................................................. ix

List of Tables................................................................................................................... xii

List of Abbreviations....................................................................................................... xv

Abstract ........................................................................................................................ xviii

Zusammenfassung .......................................................................................................... xix

Özet ................................................................................................................................. xx

1. Introduction .................................................................................................................. 1

2. Container shipping ....................................................................................................... 6

2.1 Containerization ..................................................................................................... 6

2.1.1 History ............................................................................................................. 6

2.1.2 Containers ........................................................................................................ 6

2.1.3 Vessels ............................................................................................................. 7

2.1.4 Ports ............................................................................................................... 10

2.1.5 Trade .............................................................................................................. 11

2.1.6 Land side ........................................................................................................ 14

2.2 Liner shipping ....................................................................................................... 14

2.2.1 Origination ..................................................................................................... 14

2.2.2 Shipping lines ................................................................................................ 15

2.2.3 Rates .............................................................................................................. 17

2.3 Service networks .................................................................................................. 18

3. Feeder Service ............................................................................................................ 22

3.1 Background ........................................................................................................... 22

3.2 Advantages of direct and feeder service ............................................................... 23

3.2.1 The advantages of direct service ................................................................... 23

3.2.2 The advantages of feeder service ................................................................... 24

3.3 Modern H&S service networks ............................................................................ 26

3.4 Feeder service networks ....................................................................................... 28

3.5 Feeder shipping lines ............................................................................................ 29

3.5.1 Characteristics of feeder lines ........................................................................ 29

3.5.2 Differences between trunk and feeder lines .................................................. 31

3.5.3 Effecting factors on performance of feeder lines .......................................... 33

3.6 Demand fluctuation .............................................................................................. 35

3.7 Planning levels in feeder service .......................................................................... 37

3.7.1 Strategic planning .......................................................................................... 38

3.7.2 Tactical planning ........................................................................................... 39

3.7.3 Operational planning ..................................................................................... 41

4. Literature Review ....................................................................................................... 43

4.1 Liner shipping network design ............................................................................. 43

4.2 Feeder service ....................................................................................................... 51

4.3 Vehicle routing problem ....................................................................................... 56

4.4 Container throughput estimation .......................................................................... 60

4.5 Liner shipping under unstable demand environments .......................................... 62

5. The Feeder Service Network Design Problems ......................................................... 68

5.1 The vehicle routing problem with simultaneous pickup and delivery with time

limit ....................................................................................................................... 68

5.2 The feeder containership routing problem ........................................................... 71

5.3 Feeder service network design problem ............................................................... 73

5.4 Liner shipping network design under unstable demand environments ................ 78

6. The Proposed Solution Methodology ......................................................................... 84

6.1 The adaptive neighborhood search approach ....................................................... 84

6.1.1 Saving heuristics ............................................................................................ 84

6.1.2 Variable neighborhood search ....................................................................... 86

6.1.3 Adaptive perturbation mechanism ................................................................. 89

6.2 Forecasting framework ......................................................................................... 92

6.2.1 Decomposition mechanism ............................................................................ 92

6.2.2 Estimation mechanism ................................................................................... 93

6.2.3 Simulation mechanism .................................................................................. 95

7. Numerical Investigation ............................................................................................. 96

vii

7.1 VRPSPDTL application ....................................................................................... 96

7.1.1 Benchmark instances ..................................................................................... 97

7.1.2 Numerical results ........................................................................................... 97

7.1.3 Concluding remarks ..................................................................................... 102

7.2 FCRP application ................................................................................................ 103

7.2.1 Case study .................................................................................................... 103

7.2.2 Numerical results ......................................................................................... 104

7.2.3 Concluding remarks ..................................................................................... 107

7.3 FNDP application ............................................................................................... 107

7.3.1 Implementation ............................................................................................ 108

7.3.2 Case study .................................................................................................... 109

7.3.3 Numerical results ......................................................................................... 111

7.3.3.1. Strategic options for feeder networks ................................................... 112

7.3.3.2. Demand scenarios ................................................................................. 115

7.3.4 Concluding remarks ..................................................................................... 116

7.4 LSND under unstable demand environments ..................................................... 117

7.4.1 Implementation ............................................................................................ 118

7.4.2 Case study .................................................................................................... 119

7.4.3 Demand estimation ...................................................................................... 121

7.4.4 The impact of seasonal demand fluctuations ............................................... 124

7.4.5 Experimental design .................................................................................... 126

viii

7.4.6 Numerical results ......................................................................................... 130

7.4.7 Concluding remarks ..................................................................................... 134

8. Summary .................................................................................................................. 136

A. Appendices ............................................................................................................... 140

References ..................................................................................................................... 157

List of Figures

Figure 1.1 A typical sea based intermodal container transportation chain 2

Figure 2.1: World vessel fleet by principal vessel types (millions of dwt) 8

Figure 2.2: Evaluation of containership fleet (1990-2012) 9

Figure 2.3: Global total container trade and port throughput (1996-2011) 12

Figure 2.4: Evaluation of shipping industry indexes (March 2009 - February 2013) 18

Figure 2.5: Symmetric and asymmetric line bundling networks 19

Figure 2.6: Pendulum and round the world service networks 20

Figure 2.7: Hub-and-Spoke and Interlining/Relay network 20

Figure 3.1: A modern multi layered H&S service network 27

Figure 3.2: Feeder service network as a part of H&S network 29

Figure 6.1: Structure of the ANS approach 85

Figure 6.2: Structure of construction heuristic (Kulak et al. 2011) 86

Figure 6.3: Neighborhood structures 88

Figure 6.4: Structure of APM algorithm 90

Figure 6.5: Forecasting framework 92

Figure 6.6: Typical ANN architecture 94

Figure 7.1: Improvement of the solution during the iterations (CMT13X with

service time) 99

Figure 7.2: Best solution networks for soft and hard time deadline 105

Figure 7.3: Sensitivity analysis results for penalty parameter 107

Figure 7.4: Calculation of the FNDP fitness function 108

Figure 7.5: An example of a route-ship-port schedule 109

Figure 7.6: Regional feeder and hub ports 110

Figure 7.7: Feeder route networks for Port Said (left) and Candarli port (right) 114

Figure 7.8: Calculation of the LSND fitness function 119

Figure 7.9: Regional ports 120

Figure 7.10: An example of monthly throughputs estimation 122

Figure 7.11: An example of daily throughputs simulation 123

Figure 7.12: Examples of weekly demand estimations of the Odessa container

terminal (right) and whole region (left) 124

Figure 7.13: Minimum total cost of the region for a 52-week sailing season 124

Figure 7.14: Necessary number of routes with ship types and slot capacity 125

Figure 7.15: Necessary minimum number of ships with types and capacity

utilization 126

Figure 7.16: Effect of seasonal change number on the weekly total cost of the

network (Test 1) 131

Figure 7.17: Effect of seasonal change number on the weekly capacity utilization

of the network (Test 1) 131

Figure 7.18: Impact of seasonal change number and demand assignment (Test 2) 133

Figure 7.19: Impact of demand assignment on the weekly capacity utilization of

the network (Test 2) 133

Figure 7.20: Impact of seasonal change number and owned ship number (Test 3) 134

Figure 7.21: Impact of the owned ship number and ship price (Test 4) 134

xii

List of Tables

Table 2.1: ISO standards for common container types 7

Table 2.2: Global existing and ordered containership fleet 9

Table 2.3: Containership size categories 9

Table 2.4: The top 20 world container ports in 2011 11

Table 2.5: The top 20 seaborne container exporter countries (2009-2010) 13

Table 2.6: The top 20 seaborne container importer countries (2009-2010) 13

Table 2.7: The top 20 seaborne container trade routes (2009-2010) 14

Table 2.8: The top 20 liner shipping operators in 2013 16

Table 4.1: Liner shipping service network design related studies 44

Table 4.2: Overview of feeder service network design related studies 55

Table 4.3: Overview of VRPSPDTL related studies 60

Table 4.4: Container throughput forecasting related studies 61

Table 5.1: Basic calculations of total costs for a sailing season 74

Table 7.1: Sensitivity analysis results for algorithm parameters 98

Table 7.2: The effect of the neighborhood structures (CMT13X with service time) 99

Table 7.3: Computational results for the benchmark problem instances with

service time 100

Table 7.4: Computational results for the benchmark problem instances without

service time 101

xiii

Table 7.5: Comparisons of approaches for benchmark instances 102

Table 7.6: Sensitivity analysis results for algorithm parameters 105

Table 7.7: Comparisons of algorithms 105

Table 7.8: Sensitivity analysis results for problem parameters 106

Table 7.9: Parameter values for ship types 112

Table 7.10: Scenario results for alternative hub port locations 113

Table 7.11: Feeder network comparison of the Port Said and Candarli port options 115

Table 7.12: Sensitivity analysis of market volume increase 116

Table 7.13: Extended parameter values for ship types 121

Table 7.14: An example of monthly throughputs decomposition 122

Table 7.15: Simulation coefficients for week days 123

Table 7.16: Simulation coefficients for month days 123

Table 7.17: An example of daily throughputs fluctuation 123

Table 7.18: Seasonal week allocations to periods (Scenario A) 127

Table 7.19: Seasonal demand assignment approaches (Scenario B) 127

Table 7.20: Example of seasonal demand assignments 128

Table 7.21: Owned ship number determination approaches (Scenario C) 129

Table 7.22: Example of owned ship number determination approaches 129

Table 7.23: The change in the ship prices (Scenario E) 130

Table 7.24: Designed experimental tests 130

Table A.1: The top 100 liner shipping operators in 2013 140

xiv

Table A.2: Computer / Software benchmarking results 142

Table A.3: Best solution route sequences for CMT10Y* with service time 142

Table A.4: Best solution route sequences for CMT13X* without service time 143

Table A.5: Demand, supply, service time of feeder ports and distances between

ports* 143

Table A.6: Route details of best solution for soft time deadline 144

Table A.7: Route details of best solution for hard time deadline 144

Table A.8: Demand parameters for related container terminals* 144

Table A.9: Best solution for the feeder network design of the Candarli port 145

Table A.10: Figures of contracted container terminals* 145

Table A.11: Monthly export and import rates of countries in 2005-2011 146

Table A.12: Forecasted weekly total throughputs of regional container terminals* 149

Table A.13: Results of designed experimental tests 154

List of Abbreviations

ACS

Ant Colony System

ALT

Alternating Heuristic

ANN

Artificial Neural Network

ANS

Adaptive Neighborhood Search

APM

Adaptive Perturbation Mechanism

CIH

Cluster Insertion Heuristic

cst

centiStoke, a unit of kinematic viscosity

dwt

Deadweight tonnage, ship carrying capacity measured in metric

tones

EVNS

Enhanced Variable Neighborhood Search

FCRP

Feeder Containership Routing Problem

FND

Feeder service Network Design

H&S

Hub-and-Spoke

IBH

Insertion Based Heuristic

ILS

Iterated Local Search

ISO

International Organization for Standardization

LNS

Large Neighborhood Search

LSND

Liner Shipping service Networks Design

xvi

Meter is the fundamental unit of length in the International Sys-

tem of Units

Cubic meter is derived unit of volume in the International System

of Units

MILP

Mixed-Integer Linear Programming

MLVRPSPD

Vehicle Routing Problem with Simultaneous Pickup and De-

livery with Maximum Distance Length

MPC

Multi-Port-Calling

MVRPB

Mixed Vehicle Routing Problem with Backhauls

Non-deterministic Polynomial-time

NSP

Nearest Sweep with Perturbation

PDP

Pickup and Delivery Problems

PSO

Particle Swarm Optimization

RTS

Reactive Tabu Search

Savings Algorithm

SBAA

Saving Based Ant Algorithm

TEU

Twenty-foot Equivalent Unit

Tabu Search

ULCV

Ultra Large Container Vessel

UNCTAD

The United Nations Conference on Trade and Development

USA

United States of America

US$

The United States dollar

xvii

VND

Variable Neighborhood Descend

VNS

Variable Neighborhood Search

VRP

Vehicle Routing Problem

VRPB

Vehicle Routing Problem with Backhauls

VRPMTTL

Vehicle Routing Problem with Multi Trip and Time Limit

VRPPD

Vehicle Routing Problem with Pickup and Delivery

VRPPDTW

Vehicle Routing Problem with Pick-up and Delivery with Time

Windows

VRPSPD

Vehicle Routing Problem with Simultaneous Pickup and Delivery

VRPSPDTL

Vehicle Routing Problem with Simultaneous Pickup and Delivery

with Time Limit

VRPSPDTW

Vehicle Routing Problem with Simultaneous Pickup and Delivery

with Time Windows

xviii

Abstract

In the new era of mega containerships, global containership liners design their transpor-

tation service as Hub-and-Spoke networks to improve the access to local transportation

markets and to reduce operational costs by using short-sea connections for low-volume

transportation lanes. These connections from the hub ports to the regional ports consti-

tute the feeder network which is serviced by small or medium-sized feeder container-

ships. This study analyzes general characteristics of feeder services in liner shipping and

provides operation research based solutions to major challenges that feeder service pro-

viders face in planning their service networks. For this purpose, an adaptive neighbor-

hood search approach, which is proved to be effective in vehicle routing problem vari-

ants, is developed in order to determine the feeder ship fleet size and mix, fleet deploy-

ments, service routes and voyage schedules to minimize operational costs for static and

dynamic sailing seasons. A Monte Carlo simulation and an artificial neural networks

based forecasting framework is also developed to estimate unstable throughput demands

of regional ports. In our case study investigation, we assume the feeder network design

problem of a Turkish short-sea shipping company in view of the opening of the new

Candarli port near Izmir. The cost performance of alternate feeder network configura-

tions serving the Black Sea region is compared under both stable and unstable demand

environments. Numerical results show that the new Candarli port has great potential as

hub port in the Black Sea region and feeder service network designs should consider

unstable demand environment of the regional ports.

xix

Zusammenfassung

Spätestens seit der Einführung von Mega-Containerschiffen planen Reedereien ihre

Netzwerke für die Container-Linienschifffahrt nach dem 'Hub-and-Spoke'-Prinzip, um

ihre Verbindungen zu regionalen Märkten zu stärken und die operativen Kosten für

Kurzstrecken mit niedrigem Frachtvolumen zu reduzieren. Das sogenannte Feeder-

Netzwerk besteht aus solchen Kurzstrecken zwischen Hubs und Regionalhäfen, welche

üblicherweise von kleinen oder mittelgroßen Containerschiffen bedient werden. Diese

Studie analysiert die allgemeinen Eigenschaften des Feederverkehrs in der Container-

Linienschifffahrt und schlägt OR-basierte Lösungsansätze für Netzwerkplanungspro-

bleme von Feederverkehr-Dienstleistern vor. Es wurde eine, aus der Tourenplanung

bewährte, adaptive Nachbarschaftssuche entwickelt, welche Größe, Zusammensetzung

und Einsätze der Feederflotte sowie die Routen und Reisefahrpläne bestimmt, um die

operativen Kosten zu minimieren. Außerdem wurden eine Monte-Carlo-Simulation und

ein Neuronales Netz für die Prognose und Auswertung von Bedarfen in regionalen Hä-

fen entwickelt. In einer Fallstudie wurde das Netzwerkplanungsproblem einer Türki-

schen Kurzstrecken-Schifffahrtsgesellschaft betrachtet im Hinblick auf die Eröffnung

des Hafens in Candarli. Die Performance neuer Feederverkehr-Konfigurationen für die

Schwarzmeer-Region wurde sowohl für statischen Bedarf als auch dynamische Be-

darfsentwicklungen evaluiert. Numerische Ergebnisse belegen, dass der neue Hafen in

Candarli ein großes Potential als möglicher Hub in der Schwarzmeer-Region besitzt.

Zusätzlich wird die Notwendigkeit bestätigt, dynamische Bedarfsentwicklungen bei

Planung des Feederverkehrs zu berücksichtigen.

Özet

Mega konteyner gemilerinin yeni döneminde, küresel konteyner gemi hatları yerel taşı-

ma marketlerine ulaşımlarını arttırmak ve operasyon maliyetlerini azaltmak için taşıma

hizmetlerini düşük hacimli hatlarda kısa mesafe deniz taşımacılığını kullanarak Göbek-

ve-İspit ağları şeklinde tasarlamaktadırlar. Göbek limanlar ve bölgesel limanlar arasın-

daki bu bağlantılar, küçük veya orta ölçekli konteyner gemileri tarafından hizmet veri-

len besleyici ağları oluşturmaktadır. Bu çalışma, besleyici servislerin genel karakteris-

tiklerini analiz etmekte ve besleyici servis sağlayıcılarının servis ağlarını planlamakta

karşılaştıkları temel zorluklara yöneylem araştırması temelli çözümler sunmaktadır. Bu

amaçla, besleyici gemi filosu boyutunun ve karışımının, filo yayılımının, servis rotaları-

nın ve sefer çizelgelerinin sabit ve değişken planlama sezonlarında belirlenmesi için

araç rotalama problem varyantlarında etkinliği ispatlanan bir uyarlanabilir komşuluk

araması yaklaşımı geliştirilmiştir. Ayrıca bölgesel limanların düzensiz talep çıktılarının

kestirmek için Monte Carlo benzetimi ve yapay sinir ağları temelli bir tahminleme yapı-

sı geliştirilmiştir. Vaka çalışması araştırmamızda, İzmir yakınlarında yeni açılan

Çandarlı limanı kullanılarak bir Türk kısa mesafeli deniz taşımacılığı firmasının besle-

yici ağ tasarım problemi üstlenilmiştir. Düzensiz ve düzenli talep ortamları altında Ka-

radeniz bölgesine servis vermek için farklı besleyici ağ yapılarının maliyet performans-

ları karşılaştırılmıştır. Sayısal sonuçlar göstermiştir ki yeni Çandarlı limanı bölgenin

göbek limanı olmak için büyük bir potansiyele sahiptir ve besleyici servis ağ tasarımları

bölgesel limanların dengesiz talep koşularını dikkate almalıdır.

1. Introduction 1

1. Introduction

Since the nineteenth century, the importance of global transportation has enlarged with

the strong increase in world trade. Thanks to the industrial revolution and raw material

resources, the world experienced a big increase in the international trade of goods in the

twentieth century, as freight transported from industrialized Europe to the rest of world.

However, the pattern has started to change from West to East after the World Wars.

Thanks to relatively high and cheap labor resources, Eastern emerging countries have

started to produce labor-intensive industrialized goods and transport them from East to

West. After the Cold War, Eastern emerging countries have also further developed their

economies with increased technological production capacities.

The changes in the world trade pattern have formed new global transportation net-

works. Freight is transported via a combination of transportation modes which could

include road, rail, air, and seaways, without any handling of the freight itself when

changing modes. The need to efficiently transfer the freight between these modes has

created door-to-door intermodal transport operations (commonly by using containers).

A sea-based intermodal container transport operation typically begins by picking up a

container from the sender and transferring it to a regional feeder port via truck or com-

bination of truck and train. The containers, collected from hinterland, are transferred

from feeder port to regional hub port via small-sized feeder ships on short seas. The

containers collected from regional ports are transported to hub port of destination feeder

port via large sized trunk ships on deep seas. The feeder ship transfers the containers

from a hub port to the related feeder port and then transport trucks, or a combination of

trucks and trains deliver the container to the receivers. A typical sea-based intermodal

container transportation chain is shown in Figure 1.1.

Apart from geographical limits to using single mode transportation, there are also

time and cost advantages to use multi-modal transportation over long distances. Trucks

are flexible and relatively fast and could easily reach most of the locations; however,

they have limited carrying capacity and are a bit costly. Trains could carry more goods

than trucks and are relatively cheap in cost; however, they are limited between conti-

1. Introduction 2

nents. On the other hand, ships can carry a large amount of goods at very low-cost with-

in seas and between continents but as slower speeds (see Christiansen et al. (2004) for

detailed comparison). Air modes are not considered in the content of this thesis.

Trunk

Ship

Truck

S DP

FP HP

Train

Truck

Feeder

Ship

Truck

Train

Truck

Train

Feeder

Ship

Train

Truck

S Truck RTruck

S: Shipper (sender); DP: Dry Port; FP: Feeder Port; HP: Hub Port; R: Receiver

Figure 1.1 A typical sea based intermodal container transportation chain

Seaborne shipping is the most important transportation mode in international trade.

More than 80% of the international trade in 2010 was transported overseas (UNCTAD

2012). In the shipping market, three forms of operations are distinguished: tramp ship-

ping, industrial shipping and liner shipping (Lawrence 1972). Tramp ships do not have a

fixed schedule and are used for immediate deliveries where the most profitable freight is

available. Therefore, the activities in tramp shipping are very irregular. In industrial

shipping, the cargo owner controls the ship and the objective becomes to minimize the

cost of shipping. Liner shipping, consists of ships visiting a larger number of ports with-

in a fixed route and time schedule; this is the most common transportation means where

intermodal containers on sea are concerned (Christiansen et al. 2004).

In terms of volume, the majority of the seaborne transportation is carried via tramp

and industrial shipping forms; however, more important than tonnage is total trade val-

ue. More than 70% of the total trade in terms of value is carried by the liner shipping

form (UNCTAD 2012). International merchandise trade is one of the most important

factors affecting the container shipping demand. Tandem to international merchandise

trade, total world container shipping trade increased from 28.7 million TEU (Twenty-

foot Equivalent Unit) in 1990 to 151 million TEU in 2011, and worldwide container

1. Introduction 3

port throughput has increased from 88 million TEU in 1990 to 572 million TEU in 2011

(UNCTAD 2012).

Despite the rise in the amount of containerized trade, the cost of shipping containers

(freight rates) has fallen dramatically since its initiation. Low freight rates, increasing

oil costs and the recent financial crises of the 2000’s have tremendously affected the

liner shipping industry. As a result, many of shipping lines operate their service with

margin loses ranging from -3% to -25% in 2011 (Alphaliner 2011). The decreasing

margins resulted in increased focus of the industry to redesigning service networks so

they operate more efficiently.

Parallel to the increase in containerized trade, the complexity of liner shipping ser-

vices has increased. A liner shipping carrier usually has a global service network, con-

sisting of several main (i.e. trunk) line loops between multiple continents on fixed

schedules. Liner shipping carriers have mainly two different design alternatives for their

service networks: multi-port-calling (MPC) network and Hub-and-Spoke (H&S) net-

work. In H&S networks, main ports are served usually by using mega containerships in

deep seas and feeder ports are served by using feeder containerships in short seas.

The evolution of H&S networks, particularly in minor trade routes like the Black

Sea, Africa and Latin America, is a recent popular challenge to deal with in liner ship-

ping. The expansion of demand for containerized goods has developed a growing num-

ber of ports in both national and regional markets. The growths in the containerized

trade, the global containership fleet, size of mega containerships, and number of con-

tainer ports are all results of expanding global markets.

The development of H&S networks has also given rise to the need for efficient feed-

er services. The feeder service network is comprised of ships which visit a number of

ports along predefined lines of feeder ports and feed trunk containerships as to avoid

their calling at too many ports in the region. The liner shipping feeder service network

design (FND) problem aims to find an optimal service network for a feeder liner ship-

ping service provider. In a sailing season, an optimal service network includes joint so-

lution of tactical planning decisions, such as fleet size and mix, fleet deployment, ship

routing and scheduling. The container feeder network design depends on the character-

istics of feeder ships, the feeder ship ports, the operating and chartering costs of the

ships and bunker costs, as well as container demand throughputs of the ports.

1. Introduction 4

Parallel to world trade, container throughputs have been directly affected by unex-

pected local and global crises (i.e. financial, political, etc.) as well as seasonal condi-

tions. Therefore, forecasting container throughputs of ports is playing a critical role in

all the levels of planning decisions of liner shipping lines. Since liner shipping involves

considerable capital investment and huge daily operating costs, the appropriate through-

put demand estimation of a whole sailing season will state the development of service

network design. In order to cope with the dynamic nature of shipping markets, it is im-

portant to design more agile and flexible feeder service networks.

The objective of this thesis is to provide operation research based solutions to major

challenges that feeder service providers face in planning their service networks. The

remainder of this thesis is structured as follows:

The background of containerization and details of liner shipping are presented in

Chapter 2. Chapter 3 provides information about the characteristics of liner shipping

feeder service. The compressive literature review is given in Chapter 4.

The FND problem is mathematically modeled in four levels in Chapter 5. While the

first level handles the problem in aspects of vehicle routing problem, the second level

handles the problem as feeder containership routing problem. The third level deals with

the basic FND problem of a stable sailing horizon by reducing the total transportation

cost and the last level approaches the problem more realistically by considering varying

forecasted throughput demands for a dynamic sailing season and vessel charter opera-

tions.

The first part of Chapter 6 proposes a novel solution approach (adaptive neighbor-

hood search) combined with the classic savings heuristic as initial solution construction

algorithm, variable neighborhood search in order to improve the initial solution, and a

perturbation mechanism to escape from local optima. The second part of the chapter

provides a Monte Carlo simulation and an artificial neural networks based forecasting

frame in order to analyze the impact of seasonal demand fluctuation on the liner ship-

ping feeder service.

The experimental design is presented in Chapter 7, concluding with a number of

well-known benchmark problems and a real feeder service case study from the Black

Sea region. The numerical results of benchmark studies show that the proposed method

1. Introduction 5

produces superior solutions compared to those reported in the literature and effective

feeder service networks for both static and dynamic sailing seasons. Finally, conclu-

sions are drawn and suggestions for further research are given in Chapter 8.

2. Container shipping 6

2. Container shipping

2.1 Containerization

2.1.1 History

For many thousands of years, shipping has been used to transport freight from one land

to another. Before the development of intermodal containers, break-bulk shipping was

used to transport freight from one land to another in crate, barrel and pallet forms. How-

ever, in break-bulk shipping, the loading/unloading of freight to/from ships was ex-

tremely slow and labor intensive. Since the ships were spending too much time at ports

and carrying less freight volume, shipping of freight was extremely expensive

(Levinson 2008).

The industry has developed various types and sizes of boxes for the efficient move-

ment of goods between transportation modes. These developments were too labor inten-

sive to be practical until the end of World War II. A war tanker, the Ideal X, was con-

verted with a reinforced deck to carry fifty-eight metal containers as well as 15,000 tons

of bulk petroleum by truck entrepreneur Malcolm McLean. The first voyage of it was

from Port Elizabeth, New Jersey to the Port of Houston on April 26, 1956. Please see

Levinson (2008) for evaluation of container shipping.

McLean’s intermodalism based idea aimed to move the freight with the same con-

tainer between transportation modes with minimum interruption. Intermodal containers

could be efficiently and safely transported between trucks, trains and ships. During the

years, all areas of the transport chain had to been integrated and adapted to handle the

containers in order to realize efficient intermodal container transport. This idea led to a

revolution in freight transportation and international trade over the next 50 years.

2.1.2 Containers

Mclean’s initial design for the container was at 8 feet tall, 8 feet wide and 10 feet long

units. Until the early 1960’s, there was no standardization for container constructions

and size; each shipping line was using its own standards. In 1961, International Organi-

2. Container shipping 7

zation for Standardization (ISO) set standards to help effectively transport containers

between shipping lines all over the world (Levinson 2008). Weights and dimensions of

some common types of containers are shown in Table 2.1 (Wikipedia 2013b).

Table 2.1: ISO standards for common container types

20′ container

40′ container

40′ high-cube

45′ high-cube

External length (m)

6.058

12.192

13.716

External width (m)

2.438

External height (m)

2.591

2.59

2.896

Interior length (m)

5.710

12.032

12.000

13.556

Interior width (m)

2.352

2.311

2.352

Interior height (m)

2.385

2.650

2.698

Door width (m)

2.343

2.280

2.343

Door height (m)

2.280

2.560

2.585

Box volume (m3)

33.1

67.5

75.3

86.1

Max gross weight (kg)

30,400

30,848

30,400

Empty box weight (kg)

2,200

3,800

3,900

4,800

Net load weight (kg)

28,200

26,600

26,580

25,600

Source: Wikipedia (2013b)

Standard containers are also identified as general dry purpose containers. In addition

to standard containers, there are also a range of special container types such as open top,

open side, flat rack, refrigerated, tank, etc. Open top containers are generally used for

easy loading of odd sized goods such as logs and machinery. Open side containers are

generally used for air needed goods such as onions and potatoes. Flat racks are open

side and top containers used for transportation of extraordinary sized goods such as

boats and industrial equipment. Refrigerated containers (reefers) can control tempera-

tures and allow transportation of perishable goods such as meat, fruit, vegetables, dairy

products, chemicals and drugs. Tank containers are used for transportation of liquid

bulks such as chemicals, wine and vegetable oil.

There are more than 20 million container units which equal more than 31.25 million

TEU in the container fleet including all these types (UNCTAD 2012). Please see Levin-

son (2008) and Wikipedia (2013b) for details of container standardization.

2.1.3 Vessels

Marine vessels designed to carry intermodal containers on their hulls and decks are

called containerships. From their beginnings in 1956, the designs of containerships have

been continuously changed in order to improve efficiency. The maximum ship size has

been enlarged 9.66 times from 1,500 TEU in 1976 to 16,000 TEU in 2012; fuel effi-

ciency of 4,500 TEU sized ship has improved 35% between 1985 and 2008; carbon ef-

2. Container shipping 8

ficiency on a per-mile freight volume basis has improved 75% between a 1,500 TEU

containership build in 1976 and a modern 12,000 TEU ship built in 2007

(WorldShipping 2013b). Please see Section 2.2.3 for evaluation of freight rates.

The share of containerships in the world seaborne trade is about 12.9%, but about

70% of the total trade in terms of value is carried with containerships (UNCTAD 2012).

Figure 2.1 shows world fleet by principal vessel types during the years 1980 and 2011

in millions of dwt (UNCTAD 2012).

Figure 2.1: World vessel fleet by principal vessel types (millions of dwt)

According to Alphaliner (2010, 2013a), the number of containership fleets have

been increased almost 3 times and the fleet capacity has been increased about 8.5 times

since 1990. Thus, average ship size has increased 1.39 times from 1390 TEU in 1990 to

3307 TEU in 2012. Figure 2.2 shows the evolution of containership fleets during 1990-

2012 (Alphaliner 2010, 2013a). Containerships have been growing increasingly larger

over time. In 2012, there were about 5,000 containerships in operation with more than

16 million TEU total capacities in the industry. The sizes of newly delivered container-

ships continued to grow in 2012 and 73.12% of the new ordered containerships are

sized more than 7,500 TEU. The numbers, capacities and percentages of existing and

ordered containerships according to size ranges are shown in Table 2.2 (Alphaliner

2013a). The average age of the containership world fleet is 10.90 years and the average

age per vessel was almost twice as high at 21.9 years. 23.8 % of world vessel fleet is

between 0-4 years, 27.9% are 5-9 years, 18.3 % are 10-14 years, %17.4 are 15-19 years

and the rest %12.6 are more than years old (UNCTAD 2012).

2. Container shipping 9

Figure 2.2: Evaluation of containership fleet (1990-2012)

Table 2.2: Global existing and ordered containership fleet

Size ranges

Existing fleet

Ordered fleet

TEU

Ships

% Ships

1000 TEU

%TEU

Ships

% Ships

1000 TEU

%TEU

10000-18000

163

3.29%

2080

12.69%

120

24.74%

1656

47.82%

7500-9999

332

6.70%

2880

17.57%

20.21%

876

25.30%

5100-7499

476

9.60%

2922

17.83%

5.36%

172

4.97%

4000-5099

741

14.95%

3348

20.43%

16.08%

367

10.60%

3000-3999

291

5.87%

996

6.08%

11.13%

199

5.75%

2000-2999

674

13.60%

1716

10.47%

6.80%

2.43%

1500-1999

572

11.54%

972

5.93%

8.87%

2.19%

1000-1499

699

14.10%

819

5.00%

5.15%

0.78%

500-999

782

15.78%

581

3.55%

1.65%

0.17%

100-499

226

4.56%

0.45%

0.00%

Total

4956

100.00%

16387

100.00%

485

100.00%

3463

100.00%

Source: Alphaliner (2013)

Containerships could be categorized according to their generations, type of vessels,

given dominations or largest possible size that can pass major transit canals. Table 2.3

shows a common categorization of containerships according to their capacities. Please

see Chan and Lee (2000) and Wikipedia (2013a) for more detailed categorization.

Table 2.3: Containership size categories

Name

Capacity(TEU)

Length (m)

Beam (m)

Draft (m)

Ultra Large Container Vessel (ULCV)

14,501>

366.00≥

48.80≥

15.2≥

New panamax

10,001-14,500

365.80

48.80

15.2

Post panamax

5,101-10,000

365.80

39.8-45.6

15.2

Panamax

2,801-5,100

294.13

32.31

12.04

Feeder

1,001-2,800

200-250

23.0-30.2

11.00

Small (Barge)

≤1000

≤190.00

≤23.00

≤9.50

Source: Chan and Lee (2000) and Wikipedia (2013a)

2. Container shipping 10

Feeder containerships transport containers between transshipment ports and other

regional ports. These types of ships are often customized with gear, at least when put in

service, in order to efficiently service small ports without quay cranes. The size of feed-

er containership term depends on the application. While barge containers are used in

canal/river based systems, relatively big sized containerships are started to use in elon-

gated embayment ports with the evaluation of mega containerships. Please see Chapter

3 for details of feeder service.

2.1.4 Ports

Ports represent the places where containerships could berth and exchange their contain-

er freights between sea and land sides. Inside of the ports, the operations such as load-

ing/unloading containers to/from ship, storage, transportation, and gate movements are

managed by container terminals. Berthing time of a containership in a port depends on

the number of assigned cranes to load/unload containers to/from ship and efficiency of

container terminal operations (Notteboom 2004). A port could be operated by several

terminal operators. In intermodal container transportation chain, container terminals of

ports are the gateways between land and sea-based transport networks. Please see Kim

and Günther (2010) for more details on container terminal operations.

The evolution of container shipping has led to the categorization of container ports

into three categories: hub ports, feeder ports and trunk (main) ports (Zeng and Yang

2002). The hub ports are where container transshipments may take place between trunk

(main) and feeder containerships. Feeder ports are regional hinterland gateways linked

to over-sea ports with feeder containerships via hub ports. Trunk (main) ports are re-

gional ports called by trunk ships due to their relatively high demand volumes.

Table 2.4 shows the top twenty world container ports according to their total TEU

throughputs in 2011 (ISL 2012). Total throughputs of these twenty ports increased

157.7% between 2001 and 2011 and 47.5% of containerized seaborne trade of the world

is handled by these ports in 2011. Parallel to change in global trade pattern in last dec-

ades, ports of emerging East Asia countries are dominating global containerized sea-

borne trade. 77.20% of the total throughputs of the top twenty ports are handled by thir-

teen East Asia ports. As the largest containerized trade exporter of the world, Chinese

ports represent nine of these ports. During the last ten years, Chinese ports are contin-

ued to increase their container throughputs on average 17.83%, despite the 6.8% in-

2. Container shipping 11

crease average of other top twenty ports. The other ports are mainly on the list because

of their regional hub positions.

Table 2.4: The top 20 world container ports in 2011

Ranking

Millions TEUs

TEU % growth

2011

2001

Port

Country

2001

2010

2011

2010-11

2001-11

(5)

Shanghai

China

6.3

29.0

31.7

9.4

17.5

(2)

Singapore

15.6

26.0

29.9

15.1

6.8

(1)

Hong Kong

China

17.8

23.7

24.4

2.9

3.2

(8)

Shenzhen

China

5.1

22.3

22.6

1.0

16.1

(3)

Busan

S. Korea

8.0

14.2

16.2

14.0

7.3

(50)

Ningbo

China

1.2

13.1

14.7

12.6

28.4

(31)

Guangzhou

China

1.7

12.5

14.3

14.2

23.4

(17)

Qingdao

China

2.6

12.0

13.0

8.4

17.3

(13)

Dubai

U.A.E.

3.5

11.6

12.6

9.0

13.7

(6)

Rotterdam

Netherlands

6.1

11.1

11.9

6.6

6.9

(26)

Tianjin

China

2.0

10.1

11.6

15.0

19.1

(4)

Kaohsiung

Taiwan

7.5

9.2

9.6

5.0

2.5

(12)

Port Kelang

Malaysia

3.8

8.9

9.4

6.4

9.6

(9)

Hamburg

Germany

4.7

7.9

9.0

14.2

6.8

(11)

Antwerp

Belgium

4.2

8.5

8.7

2.3

7.5

(7)

Los Angeles

U.S.A.

5.6

7.8

7.9

1.4

3.5

(47)

Xiamen

China

1.3

5.2

6.5

24.1

17.4

(49)

Dalian

China

1.2

5.2

6.4

22.1

18.1

(10)

Long Beach

U.S.A.

4.5

6.3

6.1

-3.2

3.1

(15)

Bremen

Germany

3.0

4.9

5.9

21.0

7.1

Source: ISL (2012)

2.1.5 Trade

In today’s globalized world, almost no country could depend entirely on what it domes-

tically produces. At some different levels, most of the countries are depending on inter-

national trade which could be defined as the exchange of capital, goods and services

between the countries. As explained in previous sections, seaborne shipping is the most

efficient method of transporting bulk goods. Over 90% of international trade is carried

on the water, and in terms of value, more than 70% of the trade was transported by con-

tainerships in 2011 (UNCTAD 2012). Global container trade has enlarged 2.02 times

from around 20 million TEUs in 1996 to around 151 million TEUs in 2011. In the same

period, global total container port throughput has increased 2.65 times from around 157

million TEUs to around 573 million TEUs in 2011. These indices provide evidence as

to how container trade has become an important player in the development of globalized

world economies. Figure 2.3 shows evolution of global total seaborne container trade

and port throughput during 1996-2011 in millions TEU (UNCTAD 2012).

2. Container shipping 12

The export side of international trade is extremely dominated by East Asia coun-

tries. Table 2.5 shows the top twenty seaborne exporting countries in 2009-10 (IHS

2012). According to these throughputs, global exportation of containers is also highly

concentrated. Around 31.50% of the global seaborne trade volume was exported by

Greater China (including Taiwan and Hong Kong) to rest of the world in 2010. The top

ten exporter countries account for 62.11% and the top twenty accounts for 75.68% of

the total international export. On the hand, the import side of international trade is al-

most equally dispersed around the world, except 25.90 total shares of USA and China

(see Table 2.6). Contrary to the export side, the top ten countries imported only 48.73%

of the total international trade. Similarly to the export side, the top twenty countries im-

ported almost 75% of the total international trade.

Figure 2.3: Global total container trade and port throughput (1996-2011)

The transfer direction of the trade between origin and destination countries is re-

ferred to as an international trade route. In this route, with the origin side making the

export operations, the destination side makes the import operations.

Table 2.7 shows how these trade routes are changing around the world according to

regions or country groups (IHS 2012). As a biggest exporter and second biggest import-

er country group, Great China is extremely directing the routes. Great China oriented or

destined routes are encompassing 46.37% of the total international routes. The top ten

trade routes account for 46.37% and top twenty trade routes account for 60.37% of the

global container trade volume.

2. Container shipping 13

Table 2.5: The top 20 seaborne container exporter countries (2009-2010)

Exporter

Millions TEUs

TEU % growth

Rank

Country

2009

2010

2009-10

China

26.1

31.3

19.92

United States

10.2

11.2

9.80

Japan

4.8

5.7

18.75

South Korea

4.5

5.2

15.56

Taiwan, China

2.9

3.4

17.24

Thailand

3.0

3.4

13.33

Germany

2.6

3.0

15.38

Indonesia

2.7

3.0

11.11

Malaysia

2.2

2.5

13.64

Brazil

2.3

0.00

India

1.6

1.9

18.75

Vietnam

1.3

1.6

23.08

Saudi Arabia

1.1

1.6

45.45

Italy

1.5

1.6

6.67

Turkey

1.4

1.6

14.29

Netherlands

1.4

1.6

14.29

Canada

1.4

1.5

7.14

United Kingdom

1.4

1.5

7.14

France

1.2

1.3

8.33

Hong Kong

1.2

1.3

8.33

Total

Top 20

74.8

86.5

15.64

Total

World

99.8

114.3

14.53

Source: IHS (2012)

Table 2.6: The top 20 seaborne container importer countries (2009-2010)

Importer

Millions TEUs

TEU % growth

Rank

Country

2009

2010

2009-10

United States

15.0

17.6

17.33

China

11.2

12.0

7.14

Japan

5.4

6.1

12.96

South Korea

3.9

4.5

15.38

Germany

2.4

2.8

16.67

Other Arabian Gulf

2.3

2.7

17.39

United Kingdom

2.3

2.5

8.70

Indonesia

2.1

2.5

19.05

Taiwan

2.2

2.5

13.64

Hong Kong

2.3

2.5

8.70

Western Africa

2.5

2.4

-4.00

United Arab

Emirates

2.0

2.1

5.00

Malaysia

1.7

2.1

23.53

Thailand

1.6

2.0

25.00

Vietnam

1.8

2.0

11.11

India

1.7

2.0

17.65

Brazil

1.3

1.9

46.15

Austrailia

1.5

1.8

20.00

Italy

1.6

1.8

12.50

Netherlands

1.3

1.7

30.77

Total

Top 20

66.1

75.5

14.22

Total

World

99.7

114.3

14.64

Source: IHS (2012)

2. Container shipping 14

Table 2.7: The top 20 seaborne container trade routes (2009-2010)

Trade Route

Millions TEUs

TEU % growth

Rank

Destination

Origin

2009

2010

2009-10

United States

Greater China

7.1

8.5

19.72

European Union

Greater China

5.8

6.9

18.97

Other Asia

Greater China

4.3

5.3

23.26

Other Asia

4.5

5.0

11.11

Middle East and

Africa

European Union

3.1

3.4

9.68

Greater China

United States

3.2

3.4

6.25

Middle East and

Africa

Greater China

2.7

3.3

22.22

European Union

Other Asia

2.8

3.1

10.71

Greater China

European Union

2.9

3.1

6.90

Other Asia

European Union

2.6

2.9

11.54

Greater China

2.6

2.9

11.54

Greater China

Other Asia

2.3

2.8

21.74

Middle East and

Africa

Other Asia

2.7

0.00

United States

Other Asia

2.3

2.6

13.04

United States

Latin America

2.2

2.4

9.09

Japan

Greater China

2.1

2.4

14.29

Other Europe

Greater China

1.8

2.3

27.78

United States

European Union

1.8

2.1

16.67

Latin America

Greater China

1.6

2.0

25.00

European Union

Middle East and

Africa

1.6

1.9

18.75

Total

Top 20

60.0

69.0

15.00%

Total

World

99.7

114.3

14.64%

Source: IHS (2012)

2.1.6 Land side

Since most of the containers come to ports from land side by using trucks and trains,

efficient and timely transportation of containers from their origins will affect the per-

formance of container terminals. Actually, the continued schedule of success of global

intermodal transportation chain depends on whole effectiveness of each node. There-

fore, a disruption anywhere on one of the transportation networks could result in ship-

ment delays of the cargo (WorldShipping 2013c). Since the content of this study is to

provide solutions to challenges of sea side container transportation, please see Iannone

et al. (2007) for more detailed information on inland container transportation operations.

2.2 Liner shipping

2.2.1 Origination

Until the end of the 18th century, the ships were sailing between lands according to dai-

ly wind and weather conditions. With the successful integration of steam engines to

ships in the mid-19th century, the ships started to provide regular passenger and cargo

2. Container shipping 15

service. Before developments on intermodal container transportation in around mid-

20th century, shipping lines had commonly provided a combined service with bulk car-

go, passenger and mail (Wikipedia 2013c).

After the launch of containerships in around the mid-20th century, the world had

experienced a strong increase in exchanging containerized goods and resources between

regions. With the growing demand of containerized trade, the shipping lines commonly

transformed their fleet to fully cellular containerships. Transfer simplicity and safety of

the containers have started to meet with economic efficiency. The liner shipping indus-

try has presented this service between lands more efficiently and changed the world

trade pattern day by day (Levinson 2008).

Liner shipping is accepted as the most efficient mode for transportation of goods. A

large containership with 8,000 TEU capacities could transport more than 200,000 TEUs

in one year. In order to transport this amount, it would require using hundreds of freight

aircraft, many miles of rail cars, and fleets of trucks. The containerized transport cost of

a bicycle from Asia to Europe is about US$10, a media player is about US$1.50, a kg of

coffee is just US$0.15, and a can of beer is around US$0.01 (WorldShipping 2013a).

2.2.2 Shipping lines

A liner container shipping line operates a fleet of containerships to provide shipping

service between ports on fixed routes and schedules with regular frequencies (Windeck

2013). A shipping line has to service its customers with fixed sailing schedules in order

to make containers available to ensure loading of the containers into the ships. In liner

shipping, it could be expected that a ship will serve various ports on its route. The ne-

cessity of keeping the schedule on these ports will make the route fixed as well. A result

of a deviation from this route could be non-availability of a contracted container in an-

other port. As in public transport bus service, liner shipping service has to follow regu-

lar service frequencies in order to meet periodic demands of customers. Further analyses

on determinants of container liner shipping are recently provided by Ducruet and

Notteboom (2010).

Despite the increase on the scale of liner shipping economy and oil prices over

years, the profit ranges of the shipping line industry have decreased (see Section 2.2.3).

Therefore, in order to cope with growing demand and decreasing rates, the organiza-

2. Container shipping 16

tional structure of shipping lines have commonly reformed in order to increase effec-

tiveness of their services. While some shipping lines independently continue to operate

their services, much more of them have gone to operational collaborations (liner confer-

ences, strategic sharing/alliances, and mergers and acquisitions). Therefore, the struc-

ture and the slot share of the top twenty shipping lines have significantly changed over

the last thirty years. The top twenty shipping line operators controlled 26% in 1980,

41.6% in 1992, and 58% in 2003 of the world slot capacity (Notteboom 2004). Table

2.8 shows vessel numbers, capacities and shares of the top twenty shipping line opera-

tors in 2013 (Alphaliner 2013b). 85.79% of the world container slot share is controlled

by the top twenty shipping line operators in 2013. The top three line operators have con-

trolled 38.21% and the top ten have controlled 64.73% of the total container capacity.

See Appendices (Table A.1) for more details about the top 100 liner shipping operators

in 2013.

Table 2.8: The top 20 liner shipping operators in 2013

Shipping Line

Slot

World

Total

Average Size

Rank

Operator

TEU

Ships

TEU

APM-Maersk

2,562,353

15.56%

588

4,358

MSC

2,306,196

14.01%

475

4,855

CMA CGM

1,423,193

8.64%

420

3,389

COSCO

731,588

4.44%

159

4,601

Evergreen Line

721,571

4.38%

183

3,943

Hapag-Lloyd

648,247

3.94%

140

4,630

APL

606,865

3.69%

128

4,741

Hanjin Shipping

585,309

3.56%

112

5,226

CSCL

572,283

3.48%

139

4,117

MOL

499,893

3.04%

108

4,629

OOCL

448,051

2.72%

4,619

NYK Line

414,299

2.52%

4,361

Hamburg Süd

409,118

2.48%

101

4,051

K Line

352,106

2.14%

4,959

Yang Ming

350,646

2.13%

4,329

HMM

341,074

2.07%

5,984

Zim

320,018

1.94%

3,903

PIL

300,133

1.82%

146

2,056

UASC

277,665

1.69%

5,785

CSAV Group

254,392

1.55%

4,625

Total

Top 20

14,125,000

86%

3,285

4,300

Total

World

16,464,087

100.00%

4953

3,324

Source: Alphaliner (2013b)

2. Container shipping 17

2.2.3 Rates

During the years, the freight shipment rates of the liner shipping industry have eroded

due to economic forces. With the increase in the scale of global economy, the industry

has increased the fleet capacity to cope with international trade demand. However in

these years, parallel to world trade, industry has been directly affected by unexpected

local and global crises (i.e. financial, political etc.) as well as seasonal conditions. The

major ups and downs in the economy have caused overcapacity on the slots of the

fleets. Since the liner shipping is a highly capital-industry, when the large, expensive

networks are set up, the operators make pressure to fully utilize these unused capacities

of the ships. Since the shipping industry cannot influence the total throughputs of the

market, the shipping industry has been decreasing its freight prices in order to attract

more share from the market. As a result of the erosion in freight rates and explosion of

bunker prices, shipping lines have started to operate with very low freight revenue. This

marginal cost approach often causes direct operational losses on low demand periods

due to high fixed costs (Notteboom 2004).

Figure 2.4 shows how the bunker, freight and slot index rates fluctuated between

March 2011 and February 2013. In the figure, bunker index is the average global bunker

price for all 380-centistoke (cst) port prices (US$) for per metric ton published on

BunkerIndex (2013). During the period of time in the figure, the related bunker index

fluctuated between $267 and $753 with $568 being average. Freight index represents, as

an example, the shipment price of a container from China to Europe which fluctuated

between $413 and $1872 with an average of $1210 in the same period (ShippingChina

2013). Slot index represents the daily slot rate (US$) of a chartered containership, i.e.

the figure illustrates the slot cost of a 2500 TEU sized containership according to charter

index of VHSS (2013).

Generally, Figure 2.4 implies that both uncontrolled global economy conditions and

the explosion of bunker prices results in an unstable environment for freight shipment

rates. This instability led shipping lines to intensely concentration on their network re-

lated costs in short-term perspective.

2. Container shipping 18

Figure 2.4: Evaluation of shipping industry indexes (March 2009 - February 2013)

2.3 Service networks

The liner shipping service networks are developed in order to meet the growing demand

of shipping lines in terms of throughputs, port accessibility, shipping durations, and

service frequencies. Shipping lines implicitly have to balance the requirements of the

customers and operational cost considerations when designing their networks. Custom-

ers would demand direct services between the origin and the destination of their freights

which would create an impossible pressure on the service schedules, frequencies and

routes, and as well as the complexity of networks. On the other hand shipping lines

would like to design their service networks in order to optimize utilizations of ships, to

increase coverage of ports and to minimize transportation cost by using effectiveness of

large containerships (Zohil and Prijon 1999; Lirn et al. 2004; Ducruet and Notteboom

2012).

Shipping lines could design their service in a great variant of network patterns in or-

der to optimize their service efficiency. However, the more efficient a service network

design from the perspective of carriers, the less appropriate the service network for cus-

tomer expectations could become (Notteboom 2006). Therefore, shipping network de-

sign of each shipping line is dependent upon their offered service type and covered trade

route. In contrast to conventional shipping, bundling is one of the key components of

the liner service networks. In the liner service network design, the bundling of the con-

tainers could occur at two levels: bundling within service and bundling by linking two

or more services (Ducruet and Notteboom 2012).

2. Container shipping 19

The objective of bundling within an individual liner service is to collect containers

by serving a number of ports along the similar route patterns and time intervals (multi-

port calling). Such a line bundling service usually starts from farthest contacted port of

the region and sails to the farthest contracted port of another region by visiting a set

number of ports in the regions. A line bundling service operation could be symmetric

(see Figure 2.5a) or asymmetric (Figure 2.5b), depending on the return journey (Ducruet

and Notteboom 2012). An example of such a line bundling service might be a route

from Hamburg, London, Rotterdam, and Antwerp, ports of North Sea, to Sharjah,

Mumbai, Colombo and Chennai, ports of Arabic Sea.

Pendulum and round-the-world services are extensions of the line bundling service.

In pendulum (see Figure 2.6a), liner services usually cover more than two trade routes,

i.e., from North Sea ports via Far East ports to North Pacific ports and vice versa. In the

round-the-world service (see Figure 2.6b), the ship never turns around, it just keeps sail-

ing until it completes a circumnavigation and returns to its starting point, i.e., a ship

starting from Singapore port might follow Trans-Indian, Trans-Mediterranean, Trans-

Atlantic and Trans-Pacific routes until returning back to Singapore.

2 1

1 2

Out of market

port

Port of

call

Outgoing journey

Return journey

2 1

Out of market

port

Port of

call

Outgoing journey

Return journey

Figure 2.5: Symmetric and asymmetric line bundling networks

2. Container shipping 20

2 1

1 2

Out of market

port

Port of

call

Outgoing journey

Return journey

1 2

Out of market

port

Port of

call Round journey

A B

Figure 2.6: Pendulum and round the world service networks

1 2

Out of market

port

Port of

call

5 6

Inter conti-

nental hub

Continental

hub port

T. line (continent A)

T. line (continent B)

1Out of market

port

Trunk line – Outgoing

Hub port

Feeder

port

Feeder service (shuttle)

Trunk line - Return

A B

Figure 2.7: Hub-and-Spoke and Interlining/Relay network

Hub-and-Spoke (H&S) networks, interlining and relay are the main options to bun-

dle containers by using more than one liner service. With the growing complexity of

service networks in the mid-1990s, shipping lines established hub ports in order to make

transshipment activities in order to reply the demands of market (Ducruet and

Notteboom 2012). A hub port serves as a transshipment port and the character of this

port changes depend on service patterns. In H&S service networks, the port serves as

regional transshipment center between trunk line and feeder services (see Figure 2.7a).

In this network, export containers are first delivered from feeder ports to hub ports via

feeder services, then main liner services transports these containers to destination ports.

Similarly, import containers are dispatched from hub port to feeder ports by using feed-

er services. In interlining service, a hub port serves as continental transshipment center

2. Container shipping 21

between trunk lines, and in relay service, it serves as regional transshipment center be-

tween trunk lines (see Figure 2.7b).

Further analyses on dynamics and determinants of liner shipping networks are re-

cently provided by Lam and Yap (Lam and Yap 2011), Wilmsmeier and Notteboom

(2011), and Ducruet and Notteboom (2012).

3. Feeder Service 22

3. Feeder Service

3.1 Background

From the beginning of containerization, it was commonly believed that shuttle opera-

tions could decrease the cost of container liners (McKinsey 1967). The required shuttle

transportation could be executed by road, rail and sea feeder service modes depending

on specific situations. Rail service could be an effective inland transportation mode as

long as distance, volume and geographic conditions were appreciable. Road transporta-

tion mode could be selected in low volume short distance cases where rail service was

not provided, and sea feeder service could be preferred on relatively long distances

where geographically appreciable demand existed (Jansson and Shneerson 1982).

However, until development of modern liner shipping networks, sea based feeder

services were not preferred unless road/rail transport was impossible (i.e. to island mar-

kets) due to their extra transshipment cost and longer transit time. In the early years of

containerization, a deep sea containership was calling on a relatively large number of

various sized ports (multi-port calling). Evaluation of mega sized containerships come

with efficient transportation costs over long distances. But by visiting a number of vari-

ous sized regional ports, the ships were wasting too much time on ports (Jansson and

Shneerson 1982). Therefore, as an alternative to multi-port calling transportation by

using individual liner service, H&S based transformation networks by using two or

more liner services appeared in the industry. In this network, bigger sized container-

ships serve among the trans-shipment hub ports, and smaller sized containerships pro-

vide feeder service between hub port and the regional feeder ports. Large containerships

were able to concentrate on sea crossing operations and do not waste time on small de-

mand sized ports.

Both of the service alternatives have been criticized during the years, since direct

service based multi-port calling systems and feeder (indirect) service based H&S have

clear advantages and disadvantages (Imai et al. 2009). In next section, the advantages of

the direct service and advantages of the feeder service are provided in detail. Note here

3. Feeder Service 23

that the advantages of direct service are the disadvantages of feeder service and the ad-

vantages of feeder service are the disadvantages of direct service.

3.2 Advantages of direct and feeder service

Direct service means a service where an individual ship carries the containers from

origin port to the destination port or a group of ports without transshipment of contain-

ers from one ship to another ship via hub ports during its journey. Direct service is seen

commonly in line-bundling (multi-port calling) networks. Feeder service means a ser-

vice where containers are transported by a feeder vessel from regional port to hub port

and delivered to the final port by using main and other feeder vessels via different hub

ports. Feeder service is commonly seen in H&S networks.

3.2.1 The advantages of direct service

Main advantages of using direct service are less transit time, less additional cost, more

attractive service, more reliable service, increased shelf life, and decreased transporta-

tion damage.

Less transit time: Direct service offers reduced transit time as compared to a feeder

service via transshipment on hub port. The transit time will contain only loading and

unloading operations on ports and routing time between origin and destination port.

Therefore, there will be no waiting time on hub port for next trunk/feeder line service.

Less additional cost: Since there is no transshipment operation on hub port, there

will also be no extra transshipment cost on hub ports and feeder service cost to transfer

containers to regional ports (Cullinane 1999).

More attractive service: When direct service operators provide higher capacity ships

between potential high demanded ports, the service operator could reduce freight cost

with the help of less transit time and less additional cost. This may help individual, di-

rect service operators to attract more shippers in the market as compared to transship-

ment based service operators (Ducruet and Notteboom 2012).

More reliable service: Since the ships in direct services are not correlated with the

other ships, they are not affected by other ships impacted by delays. Any delay in feeder

service could cause missing of trunk line service which means waiting for the next trunk

3. Feeder Service 24

ship to get loaded. Generally, the gap between the two sailings would be a week and

this delay will have a major impact on the trade sustainability of shippers.

Increased shelf life: Faster transit time will come with the increased shelf life of

products. This will give a chance for traders to deal with the transportation of perishable

goods or sensitive health care products, etc. In addition, less transit time will decrease

energy consumption of refrigerated containers (reefers).

Decreased transportation damage: Another advantage of using direct service is that

there is less risk of damage, since the container is handled less often than feeder service

based transportation during its journey.

3.2.2 The advantages of feeder service

Feeder service based networks are selected by global shipping operators due to the fol-

lowing advantages.

Increased port range: Trading of goods is not limited to any specific region or a

specific port of the region. In today’s world, every region has its own specific produc-

tion potential and the excess of production beyond their consumption could be sent to a

different region where the demand exists or insufficient products could be supplied

from a different region where the excess of related product exists. Therefore the demand

of the different regions and different ports of a region is dependent on the requirements.

The demand of the small sized ports cannot be met with economical requirements of

large sized ships (Jansson and Shneerson 1982). The feeder service allows these small

sized ports to meet with the rest of the world; shipping line could be able to cover a

range of ports around the service networks.

Eliminates port restrictions: Serving mega-sized containerships presents several

problems for small sized ports which have restrictions on berth draft and lack of ade-

quate handling equipment. Therefore, in order to benefit from the increasing efficiency

of mega-sized ships, these types of small ports could be served with feeder ships.

Increased benefit from small sized ships: The demand of regions and ports of a re-

gion is different. The low potential demand of extra small ports does not mean that they

will not be covered by global networks. The barge and feeder sized ships are still quite

efficient under low demand conditions in both short and long distances compared to

3. Feeder Service 25

mega containerships. Using these types of ships in feeder services, extra small sized

ships could become a part of global transportation networks.

Increased service number: When a large containership is deployed between conti-

nents, generally it takes at least thirty-five to forty days to complete a voyage. Since

liner shipping requires fixed routes and schedules with regular frequencies, it is neces-

sary to deploy a number of ships. Putting operation to such a large number of ships

could be covered only with adequate demand from the ports. However, this could al-

ways not be satisfied by covering a limited number of ports directly. Economical effec-

tiveness of large containerships comes from capacity utilization. In order to reach nec-

essary utilization, operators of the direct shipping service increase the interval between

service frequencies to load more containers. Increased service intervals will melt away

the attractiveness of direct service, which comes from less transit time. On the other

hand, feeder services will carry demand and supply of a range of regional ports to trunk

line ships. In order to cover this demand, shipping lines will increase the number of ser-

vice numbers. The time interval between services and waiting times in hub ports will

decrease.

Increased benefit from mega containerships: By calling on a fewer number of ports

with high demand volume, mega containers could concentrate on long distance sea

crossing operations (Imai et al. 2009). Requested benefit of mega-containerships could

be handled with only high capacity utilization which decreases the related capital cost of

per transported container.

Decreased network cost: Despite the related transshipment and transfer cost of feed-

er service based systems, increased demand of hub ports, increased service numbers and

increased benefit from mega containerships could decrease overall network costs of

global shipping lines (Imai et al. 2009).

Decreased inland traffic and air-pollution: Among the other advantages of feeder

services, the concentrated sea based network will decrease inland freight traffic conges-

tion and air pollution problems caused by road transportation (Liao et al. 2010).

In addition to main advantages of feeder service, efficient distribution of containers

to far away regions through feeder service, out of main line regions could be able to

subsist in a worldwide market.

3. Feeder Service 26

3.3 Modern H&S service networks

In the beginning of implementation of hub ports to service networks, the shipping in-

dustry was curious about the cost efficiency of the system. Economic scale of relatively

small sized trunk ships was not sufficient to cover extra transshipment cost and feeder

service cost (Lun et al. 2010). Day by day, increasing size and efficiency of large scale

containerships are converted from hub ports to essential nodes of almost all service

network patterns by maximizing port coverage and minimizing total transportation cost.

Although different in scale, relatively huge sized regional ports act as hub ports to other

small sized regional ports with feeder services in almost all of the service patterns.

While this feeder service is more complex in H&S spoke networks, it is also somehow

existing and critically important in other network patterns.

Traditional H&S networks (see Figure 2.7a), widely used in early ages of trans-

shipment operations, was originated from airline transportation. The common aim was

collection and distribution of containers from regional ports to hubs with direct shuttle

services, and transshipment of containers among hub two ports via relatively bigger

sized ships (Lu and Meng 2011).

Increasing requirements of both shipper and operators led to development of the

modern H&S service networks. Thus, less transit time and more service number de-

mands of shippers are met with sustainable and cost-effective design demands of opera-

tors in modern service networks (Løfstedt et al. 2010). A global service network fre-

quently and reliably connects feeder ports with hub ports and the main ports of the re-

gions by merging effective sides of both direct and feeder services in order to increase

competitiveness of network for specific situations. Figure 3.1 shows a modern service

network design which merges pendulum line bundling services with efficiency of H&S

services. An example of such a multilayered service might be a route using a mega con-

tainership from Hamburg to Busan by using Rotterdam and Le Havre as trunk ports, by

using Algeciras and Port Said as a hub port in Mediterranean, and by using Singapore

and Hong Kong as a hub port in East Asia.

3. Feeder Service 27

1 2

Out of market

port

Trunk

port

Trunk line – Outgoing

Hub port Feeder

port Feeder service (cyclic)

Trunk line - Return

Figure 3.1: A modern multi layered H&S service network

The evolution of H&S networks has led to hierarchical categorization of container

ports into three categories: hub ports, feeder ports and trunk ports (Zeng and Yang

2002). The hub ports are where container transshipments may take place between trunk

and feeder line containerships. The hub ports have commonly high productivity ratios

on loading and unloading of container to trunk and feeder ships. Feeder ports are re-

gional hinterland gateways linked to over sea ports with feeder line containerships via

hub ports. Due to both their geographical location, technological and low productivity

limitations, feeder ports are commonly not visited by trunk line containerships. Trunk

(main) ports are regional ports called by trunk lines due to their relatively high demand

volumes. Trunk ports usually have medium to high productivity, favorable geographical

location, and relatively good inland connections.

In the liner shipping industry, there is no fit to all approaches for hierarchical port of

call position of a container port. The port hierarchy is determined by the strategic, tacti-

cal and operational planning level decisions of individual shipping lines. The hierar-

chical ports of call decisions of these lines are rarely identical for whole liner shipping

industry. Therefore, a port may operate as a feeder port for a shipping line and a trans-

shipment hub for another line. Alternatively, a shipping line might benefit from a hub

port of another line as a trunk port.

3. Feeder Service 28

3.4 Feeder service networks

Modern global H&S networks are led to two design challenges: trunk line design and

feeder service design. Generally, trunk line design determines which ports will be used

as a hub port and it provides line bundling based on route sequence for called hub and

trunk ports. In addition, it could provide detailed information about service frequencies,

routes, schedules, deployed numbers of fleet mix in these routes, and some additional

management challenges (see Section 3.7).

In addition to network design of a trunk line, design of regional feeder service is a

critical issue in designing whole global H&S networks of shipping lines. Because the

regional ports do not have enough cargo demands to fill ships, they cannot attract the

main lines to operate a regular service. The feeder services allow these ports to meet

with the world. In conceptual terms, the feeder service is meant to simultaneously col-

lect/distribute containers from/to specific regions with feeder ships and feed/discharge

trunk containerships at hub ports as to avoid their calling at too many regional ports.

The connections between hub port and regional ports could use a shuttle feeder service

containing one feeder port or a cyclic line bundling service by containing more than one

feeder ports (Wijnolst et al. 2000). The first service strategy has the lowest transit time

but typically requires more feeder ships and smaller feeder containerships. In contrast,

indirect feeder services benefit from economies of bigger ship size but incur longer dis-

tances and longer transit times. Figure 3.2 represents a feeder service network design as

a part of H&S network. The majority of massive feeder service networks are located in

the zone of landlocked seas or huge sea gulfs (Jadrijević and Tomašević 2011). Exam-

ples of such a network could be seen in the East Mediterranean area which covers Black

Sea region ports, Sea of Marmara region ports, Aegean Sea region ports and East Medi-

terranean Sea region ports via Port Said. The various sized feeder containerships could

serve these regional ports with both shuttle and cyclic service routes (Polat et al. 2012).

Feeder services play an irreplaceable role in global shipping networks (see Section

3.2.2 for advantages of feeder service). It was the feeder service network design that

made the entire container service economically rational, efficient and more profitable,

and consequently cheaper and timely for the end users (Rudić and Hlača 2005).

3. Feeder Service 29

Hub port

Feeder port Feeder service - Cyclic

Feeder service - Shuttle

Figure 3.2: Feeder service network as a part of H&S network

The feeder service design could be regarded as a typical vehicle routing problem

variation (Andersen 2010). It deals with simultaneous transportation problems of feeder

lines in order to pick up containers from feeder ports to hub port and deliver containers

from hub port to feeder ports. In this problem, feeder lines commonly aim to design

optimal service routes by using a fleet of capacitated heterogeneous feeder container-

ships under ship due date constraints for returning to the hub port at minimum cost.

With these specifications feeder, service network design problems fundamentally fit to

the vehicle routing problem with simultaneous pickup and delivery with time limit

(Polat et al. 2012). See Section 5.3 for details of feeder service network design problem.

3.5 Feeder shipping lines

3.5.1 Characteristics of feeder lines

Although a single shipping line could operate both trunk and feeder service, it is in-

creasingly common that regional shipping lines provide feeder service in short seas for

global shipping lines (Andersen 2010). Depending on the size of regional economies,

global shipping lines could use their own subsidiary feeder line services or third party

feeder line services (Foschi 2003).

As long as enough demand cached from the region, global shipping lines are usually

operating their own subsidiary feeder service lines which are only responsible transfer

containers between destination ports and hub ports (Styhre 2010). The subsidiary feeder

lines operate together with trunk lines to catch more freight from the shipper market.

3. Feeder Service 30

Operators aim to decrease total network cost by serving related regional demands of all

vessels of a trunk shipping line. The third party common feeder lines are usually used

for low regional demands by global shipping lines. This type of operator aims to max-

imize total network revenue by serving a number of trunk shipping lines. They allocate

slot space in ships to many global shipping line customers and charge their customer on

the basis of total slot usage per voyage. Since the customers of the third party operators

are usually global shipping lines, they do not get into competition with them in freight

market.

For global shipping lines both feeder service operation decisions have the same ad-

vantages. The main advantages of using subsidiary service are low freight costs on high

volume demands, more flexible feeder vessel schedules, full control on slots of feeder

ships, and more flexible service networks. The advantages of using third party services

are sharing of operating cost with the other customers, paying the cost of only used

slots, no pressure to increase utilization of feeder ship, more frequency feeder services,

less transit times, and no competition with regional shipping lines.

The top twenty shipping lines, which control 85.79% of the world container slot and

66.50% of the total fleet size, could be defined as global shipping lines (Alphaliner

2013b). Within global lines, the top three shipping lines, which control 38.21% of world

slot and 29.98% of total fleet, are operating their feeder service almost with its own

fleet. The other global shipping lines are benefiting from both owned feeder service and

third party services depending on the conditions of the regions. Except these twenty

shipping line, structure of the top 100 shipping lines are in a great variety which oper-

ates 97.26% of total world slot (see Table A.1 in Appendices). Some of them just con-

centrate on the trunk line operations between region with couple of mega containerships

and benefitting from common feeder shipping lines for their regional services. A few of

the top fifty shipping lines are large feeder shipping lines which totally concentrate on

common feeder service with vast fleets in different regions. Within the top 100, the top

fifty shipping lines operate almost 95% of the total world slot. The rest fifty shipping

lines are usually small fleet sized direct shipping operators or small ship sized regional

feeder service operators.

3. Feeder Service 31

3.5.2 Differences between trunk and feeder lines

The main concepts, components and challenges are generally the same for trunk and

feeder lines (Andersen 2010). The significant differences between feeder and trunk

shipping lines could be compared as follows:

Operation area: The trunk lines operate in deep seas between regions and feeder

lines operate in short sea within a region. While the trunk lines usually cover global

service network, feeder trunk lines have limited regional service networks.

Demand volume: Since trunk lines serve strong hinterland connected main ports and

regional transshipment hub ports, the demand volume for transportation is very high.

On the other hand, since feeder lines serve relatively small regional ports, the demand

for transportation is very low.

Vessel size: The high demand volumes of main and hub ports and long distances be-

tween regions allow trunk lines to benefit from the effectiveness of mega container-

ships. The demand volume and operating scale of short sea shipping make it necessary

to operate with small sized ships in feeder service. Please see Sys et al. (2008) for more

details about scale economies of containership sizes and operations.

Service frequency: In liner shipping, it is expected to serve each port at least one

time in each week to meet customer demands and to provide customers with a regular

schedule. However high demand volume trunk lines usually increase service frequen-

cies of hub ports in particular. In order to optimize operation costs, subsidiary feeder

lines generally operate less frequent service to feeder ports. And, third party common

feeder lines act like trunk lines in service frequency, since they serve generally more

than one trunk lines.

Voyage time: Parallel to size of ships, the loading and unloading times of trunk line

ships on the ports are longer than port operation times of feeder line ships. In addition,

parallel to distance between regions, the trunk lines need more time to cross seas despite

higher operation speeds of mega ships. On contrary, the total voyage times of feeder

lines are rather narrow, due to short distance and low port operation times.

Fleet size: Depending on the voyage time, service frequency, and the covered re-

gions, trunk lines usually need medium/large fleet sizes in order to meet necessities of

3. Feeder Service 32

their complex service networks. On the other hand, related to short voyage times and

less service frequencies, feeder lines need small/medium fleet size in regional service

network.

Demand pattern: Main ports are usually localized in strong industrialized hinter-

lands and hub ports have wide connections with various regional ports. Therefore, trunk

lines have usually less affected from seasonal demand fluctuations. On the other hand,

the demand patterns of regional ports are rather unstable and seasonal. Hence, feeder

lines limited scales; they have rather affected from seasonal demand fluctuations. See

Section 3.6 for more information about the effect of demand fluctuation on operations

of feeder lines.

Planning horizon: Trunk lines are more restricted to their service network; usually

they plan their operations for medium to long term periods, since their huge capital in-

vestments. On the other hand, feeder lines are more flexible to adapt their self to chang-

es on market environment.

Fleet ownership: Shipping lines could be owner of operated ships, or they could

charter them as for a voyage time, or monthly, seasonally, yearly etc. Since trunk lines

have operated on more restricted network pattern, they predominantly operate with their

own ships which decrease costs over long term periods. Feeder lines usually operate a

small, fixed number of owned ships and balance its requirements with chartered ships.

They could decrease their capital costs and make their network more flexible to changes

in trade.

Slot capacity: While trunk lines work with fixed slot capacity during the planning

horizon, with the help of low chartering costs of small ships, feeder lines could operate

with flexible carrying slot capacity.

Service schedule: Trunk lines operate under fixed service schedules for defined

planning horizons. On the other hand, feeder lines could change their schedules a num-

ber of times in a planning horizon. With the help of flexible schedules, feeder lines

could adopt themselves to seasonal demand fluctuations.

Service strategy: Trunk lines depend on the demand on their market coverage could

provide direct of transshipment based service. Feeder lines generally provide direct ser-

3. Feeder Service 33

vice to customers. However, for some far away regions some large sized feeder lines

could use another small sized regional feeder service, as well.

Customer type: While the customers of trunk lines are shippers; customers of feeder

lines are global shipping lines.

Port selection: There are some common and unique factors in port choice behaviors

of trunk liners and feeder service providers. Local cargo volume, terminal handling

charge, land connection, service reliability and port location are most common im-

portant factors for trunk and feeder service. On the trunk liners side, water draft, feeder

connection, and port due are also determining factors. On the other hand, berth availa-

bility, transshipment volume and cargo proﬁtability are the other determining factor for

feeder service providers (Chang et al. 2008).

Collaboration and competition: There is high collaboration required between trunk

and feeder lines in order to create efficient service networks. Since the trunk lines are

the cheapest mode to transfer containers across oceans, they usually have only competi-

tion with other trunk lines. On the other hand, the feeder lines generally compete with

direct lines and other shipping lines, as well as regional truck and rail operators.

In conclusion, feeder lines are the intermediaries of the complex service networks

between regional shippers and trunk lines. While trunk lines connect main global ports

to each other; feeder lines help secondary ports, which have irregular and low quanti-

ties, to survive.

3.5.3 Effecting factors on performance of feeder lines

Performances of feeder shipping lines are affected from by various factors. These fac-

tors could be mainly categorized as external and internal factors. Market, customer, port

and surrounding factors are external factors and management and vessel factors are in-

ternal factors (Styhre 2010).

Market factors: The numbers of refrigerated, dangerous and standard containers, the

imbalance of import and export containers, the mix of full and empty containers, daily

and seasonal demand fluctuations, and the competition and cooperation with other re-

gional feeder lines are affecting factors on the design networks and as well as perfor-

mance of shipping lines.

3. Feeder Service 34

The numbers of refrigerated and standard containers will define transportation ca-

pacities of vessels due to essential power requirements of refrigerated containers and

limited power supply slot of ships as well as limited dangerous container stacking area.

The imbalance of import and export trade of region ports will configure the se-

quence of the ports; the vessels commonly at first will serve import intense ports and

then will serve export ports in order to maximize transportation volume. Other effecting

factor of trade imbalance is the cargo mix of empty and full containers.

The daily and seasonal demand fluctuations of regional ports have critical im-

portance on the configuration of all planning decisions of feeder service networks; be-

cause the mix and number of ship fleets, deployment of owned and chartered ships, the

sequence of ports etc. will be planned according to demand forecasts of the regions (see

Section 3.6).

Since feeder shipping lines operate under low freight rates, competition and cooper-

ation of shipping lines and inland transportation modes are also critical on the perfor-

mance of feeder services.

Customer factors: The main customers of feeder service lines are usually global

trunk shipping lines. Usually schedules of feeder ships are planned according to arrival

and departures of trunk ships. The feeder vessels have to follow schedules of trunk ships

in order to decrease waiting times of containers in hub ports. The waiting time will both

increase stacking cost of a container in the yard area of container terminal and transit

time of a container from origin to destination.

The delays of trunk ships are also important factors on stability of feeder service. In

addition, efficient information exchange between shipping lines will affect stowage

planning decisions of feeder lines.

Port factors: Compared to hub ports, the equipment infrastructures of container ter-

minals and quay depths of feeder ports are quite scarce and in wide variation. Therefore,

loading and unloading turnover durations are higher in feeder ports. In addition to these

specifications, working hours, additional pilotage requirements on berthing, bunker and

cleaning facilities of feeder ports are significant on design of networks. On the other

hand, despite usually huge infrastructures, hub port gives priorities to trunk ships which

could also effect of overall performance of feeder services.

3. Feeder Service 35

Surrounding factors: Long queue times at both feeder and hub ports, weather condi-

tions on the sea and ports, regional safety, security and environmental legislations and

regulations are affecting factors on performance of feeder services. Long queue times

and weather conditions usually create delays on route schedules and increases bunker

consumptions. Since the feeder containerships usually operating in short sea areas, they

are more restricted to use high quality bunkers in order to decrease emotions (Wang et

al. 2013b; Windeck 2013). Since the feeder containerships usually operating in short sea

areas, they are more restricted to use high quality bunkers in order to decrease emis-

sions.

Management factors: Organizational structures, efficient decision support tools, and

size and mix of owned fleets are internal factors affecting performance of feeder lines.

The third party or subsidiary company role of feeder line for a trunk liner will affect

property of encountered planning and organizing problems (see Section 3.5.1). Owned

computer-based decision support tools will also help to deploy efficient solutions to

faced planning problems such as route design, vessel stowage planning, scheduling etc.

The size and mix of owned fleets will allow great flexibility in handling fluctuations on

both regional and trunk ship operations.

Vessel factors: Another affecting factor is the specifications of owned or chartered

ships such as numbers, capacities, lengths, beams, draft, speeds, ages, geared equip-

ment, electrical power supplies, charter and purchasing costs, and bunker consumptions.

The vessel related factors have intensive influence on all planning level decisions of

feeder lines.

3.6 Demand fluctuation

The demand for liner shipping is generally closely linked to the development of world

economy and world trade (Zachcial and Lemper 2006). There are also nearly coopera-

tive relations between regional economic developments and feeder services. On the one

hand, regional economic development affects the supply of export goods as well as the

demand of import goods and raw materials which are the need of global liner shipping.

On the other hand, efficiency of feeder service in H&S system allows world-wide eco-

nomic exchange of goods.

3. Feeder Service 36

Liner shipping as well as feeder service requires high capital investment, because of

the huge capital of fixed and variable costs of containerships. The return of these in-

vestments depends on transported container volume. Therefore, a change in world or

regional trade will lead to a change in transportation volume (Lun et al. 2010). In addi-

tion to volume, the balance between import and export volume of ports will affect the

revenue of shipping service. Theoretically, a feeder ship could carry up to twice of its

slot capacity in a cyclic route. It will depart from the hub port with the import contain-

ers, will deliver the import container to regional feeder ports, will simultaneously pick

up export containers from them, and arrive back to hub port with export containers.

When the trade is imbalanced in the ports, some slots could become idle in the depar-

ture or arrival of the ship from/to hub port. The idle slots will be more if there is an im-

balance in the trade of the whole region and will be less if there is a balance in the relat-

ed region. The idle slots will effect utilization of ships; a decrease in utilization will

cause an increase in the total transportation cost per container.

The demand for liner shipping fluctuates over a year with seasonal changes, peaks at

certain times of years, and unexpected sharp drops and cancelations (Schulze and Prinz

2009; Polat and Uslu 2010). The production and consumption of some goods could vary

over the year, some following harvest seasons for fruit or fish and others following pub-

lic, national, and religious holidays. While some of these are affecting a single port or

region, several of them could create peaks in global trade, like Christmas and Chinese

New Year. Another affect which causes demand fluctuation is unexpected local and

global crises periods (i.e. financial, political etc.). In these periods, global or regional

liner shipping industry could usually experience a sharp decline in demand. In addition

in liner shipping, shippers usually pay for container transportation when the container is

loaded into a vessel or delivered to its destination. This situation allows shippers to can-

cel their bookings before loading, even their long term contractual agreements. Hence

the demand of the ports is occasionally steady during a year (Løfstedt et al. 2010).

The demand of ports reflects the necessary slot capacity for a liner shipping line.

Since the demand is uncertain, shipping lines must carefully consider their capacity de-

cisions on whether or not to expand it. However, postponing the increase of slot capaci-

ty could lead shipping lines to the risk of carrying less than capacity when the demand

volume is enlarged (Lun et al. 2010). In addition to capacity decision, the demand is the

3. Feeder Service 37

driving force in the design of service network; even small variations of the demand pat-

tern could prompt to entirely different service network designs (Andersen 2010).

Since accurately predicting the condition’s effect on liner shipping is almost impos-

sible, making reliable forecasts with certainty is also nearly impossible. But that does

not mean forecasting is pointless. The aim of forecasting is not to estimate accurately, it

attempts to help decision-makers to understand the future by reducing uncertainty by

exploring the current information. Therefore, forecasting container throughputs of ports

is playing a critical role in the planning decisions of liner shipping lines. Since liner

shipping involves considerable capital investments and huge daily operating costs, the

appropriate liner shipping feeder service network design will affect the development of

the feeder shipping lines.

Under conditions of high uncertainty, planning methods are usually based on deter-

ministic forecasts, which may be prone to failure in the long run. More realistic stochas-

tic forecasting methods, known from the academic literature, are not preferred in liner

shipping because of their complexity and high statistical data requirement (see Chapter

4). On the other hand, simulation could be used to assist with constructing a forecasting

frame by using deterministic forecasting methods that only need limited data. Indeed, a

simulation-based forecasting frame might be better suited in a stochastic environment

where unexpected drops or peaks could occur.

The dynamic, complex, and flexible nature of feeder service makes accurate fore-

casting a long-term challenge for feeder lines. Therefore, it is important to develop an

efficient methodology for forecasting container demands in order to better assist feeder

line companies in developing strategies and investment plans (see Section 6.2).

3.7 Planning levels in feeder service

As in liner shipping, decisions in feeder service are commonly characterized under stra-

tegic, tactical, and operational planning levels. The main challenges are generally same

for liner shipping and feeder services. Please see Christiansen et al. (2004), Christiansen

et al. (2007), Andersen (2010), and Windeck (2013) for more detailed information about

problems faced in strategic, tactical and operational planning levels in liner shipping.

3. Feeder Service 38

The planning decisions in feeder service depend on the organizational structure of

feeder lines. If the related feeder line is a subsidiary firm of a global shipping line, the

planning activities of feeder services will be more dependent to future plans of global

line. On the other hand, the activities of the third party common feeder lines will be

more relevant to expectations from shipping market.

Although the problems are presented below in a certain planning level, some of

them might span to more than one planning level and/or might contain collaborative

decisions with trunk lines. Since this study covers the challenges in the feeder service

network design, other decision problems in land and port side operations are not exam-

ined in this study.

3.7.1 Strategic planning

Strategic planning levels include long term decisions which are taken by top manage-

ment of feeder services. In liner shipping, while long term strategic decisions refer to

one to five years for trunk lines, it usually refers to one to three years for feeder lines

(Andersen 2010). In some long term projects such as new building terminals or fleets,

this period could spread over five to ten years. Main strategic planning decisions for a

feeder line are generally selection of service region, selection of feeder ports, hub port

options, ship types, firm scale, and ownership of fleet.

Selection of service region: A subsidiary feeder line will serve sub-regions of a

trunk line’s transshipment hub port. Hence, service region selection of subsidiary firms

are related to hub port selection of trunk lines. A subsidiary firm of a global line could

operate in only one region or could operate more than one region served by a trunk line.

When trunk lines do not use a subsidiary feeder line in the region of a transshipment

port, they have to link feeder ports by using third party feeder lines. These common

feeder lines select their service markets according to current and future development

and competition expectations of regions.

Strategic options for hub ports: Considering the economic progress in the region and

the prospects of international trade relationships as well as trends in the choice of the

transportation mode, scenarios reflecting the future development of demand for contain-

er traffic between the regional ports have to be defined. These scenarios are used to

3. Feeder Service 39

evaluate different feeder network configurations, in particular, the strategic options for

hub ports.

Selection of feeder ports: Subsidiary feeder lines do not have the chance to select

feeder ports; they have to serve all regional feeder ports which have demands and/or

supplies to trunk shipping line by minimizing network transportation costs. On the other

hand, common carriers might not provide services to some low demand and/or supply

volume ports, since they try to maximize the revenue from the region.

Ship types: Selection of service region and feeder ports comes with alternate feeder

ship types. The length, breadth, and drought of a ship could not be feasible for sailing in

the region, passing from straights or canals and approaching to ports. Moreover, termi-

nal specifications of feeder ports are also important in the selection of alternative feeder

ship types. Hence, feeder lines have to design service network according to specific

requirement of regions (see Chapter 5).

Firm scale: The slot capacity of a feeder line with operation scale affects the per-

formance of the firm. The total carrying capacity of feeder lines reflects the characteris-

tic of operation in the regional market. Large slot sized firms are more likely to create

high freight rate pressure and market share against their competitors.

Ownership of fleet: Another strategic decision on feeder lines is the ratio between

owned and chartered ships in the fleet. Feeder lines usually operate a small fixed num-

ber of owned ships and balance its requirements with chartered ships. This could de-

crease their capital costs and make their network more flexible to changes in trade.

However, if there is a stable or increasing demand trend in the market, operating with a

high number of charter ships could be couple of times more costly than operating with

owned ships. Therefore, it is crucial for feeder lines to define the minimum number of

owned feeder ships for long term efficiency (see Chapter 5).

3.7.2 Tactical planning

Tactical planning levels usually include medium term decisions which are taken by

transportation planning departments of feeder lines. In liner shipping, tactical planning

levels focus on planning decisions which take from 2-3 months up to 1 year. The tacti-

cal level problems usually consist of decisions over designing of feeder service net-

works. Main tactical planning decisions for a feeder line are generally contract man-

3. Feeder Service 40

agement, service frequency, ship routes, ship scheduling, fleet size and mix, and fleet

deployment.

Contract management: A key decision problem at the tactical level is contract man-

agement which involves the analysis and the development of the existing contract rela-

tionships with ports in the region and with cooperating companies.

Service frequency: In feeder service, it is expected to serve each feeder port at least

one time in each week to maintain customer demands and to provide customers with a

regular schedule. However, feeder lines could change service frequencies to ports ac-

cording to their demand volumes. Feeder lines could increase service frequency for

highly demanded feeder ports in order to increase satisfaction of shippers and decrease

low demanded feeder ports in order to decrease total transportation cost. Therefore, it is

important for feeder lines to decide service frequencies of feeder ports (see Chapter 5).

Ship routing: This problem aims to construct optimal service routes for a fleet of

vessels by defining service sequences of a set of ports which have both pick-up and de-

livery containers. Each feeder port has to be served once for both operations with a giv-

en fleet of identical capacitated feeder ships. Each ship leaves the hub port carrying the

total amount of containers it has to deliver and returns to the hub port carrying the total

amount of containers it must pick-up. While subsidiary lines aim to minimize total

transportation costs by satisfying all demands of related feeder ports, common lines aim

to maximize the profit from the region by sometimes declining some low profit ports.

Fleet size, mix and deployment: The aim of the fleet size and mix decision is to de-

termine the optimal composition of the fleet. The characteristic of the containerships are

important to calculate operational costs of ships. For a feeder line, operational costs for

ships include fixed costs (owning, chartering, operating, management, insurance, etc.)

and variable costs (on-sea bunker cost, on-port bunker cost, port set-up charges). Fleet

deployment is the allocation of the most suitable ship types to specific routes. In feeder

service, a deployed ship will be available after unloading whole pick up containers in a

hub port. Since the feeder ports have to be served with defined service frequencies, the

interval duration between two service times could not be enough for returning back to

hub port. Therefore, it is usually necessary to deploy more than one ship to routes. The

total fleet size and mix is the summary of all deployed ships to all routes of service net-

work. Some of the deployed fleet could be owned ships; some of necessary ships could

3. Feeder Service 41

be chartered in from the market. Since the high capital fixed and variable costs of ships,

if there are unnecessary or over capacity ships they could be chartered out.

Ship scheduling: Even scheduling is not a common problem in transportation; liner

shipping has essential features that make scheduling decisions an integral part of its

network design. Ship scheduling is one of the most essential and nonetheless most prob-

lematic planning problems for feeder lines. In feeder service design, in a given set of

port sequences (routes) and available time-windows of ports, scheduling mainly con-

cerns with the appointments and arrival and departure times to deployed ships and de-

termination of expected berthing time. In feeder service, the deployed vessels usually

lack of schedules due to congestion at ports, delays in berthing, delays in trunk lines,

inefficient terminal equipment, worse weather conditions, waiting for tug and pilots,

accidental delays and channel/straight queues (Varbanova 2011a).

The routing, scheduling and fleet size, mix and deployment problems are main parts

of the feeder service network design. The decisions made in any of these problems af-

fect the decisions in the other problems as well. As an example, even if the routes are

optimally designed, a poorly designed ship schedule could increase the necessary num-

ber of ships and decrease total profit of the network. Therefore, an efficient feeder ser-

vice network design requires simultaneous solutions to these problems. In feeder service

network design problems, all decisions related to three individual problems have to be

given at same time (see Chapter 5).

3.7.3 Operational planning

Operational planning level includes short terms decisions which could span from a few

hours to a few months. The operational planning level decisions are usually linked to

decisions made at tactical or strategic level. Main operational planning decisions for a

feeder line generally determine sailing speeds of deployed vessels, vessel stowage plan-

ning, environmental routing, and empty container repositioning.

Sailing speed: Sailing speed concerns the optimal average speed between two ports

in route sequence or average speed of the voyage. Since a 30% decrease in sailing speed

of vessels reduces the fuel consumption by 50% and greenhouse gas emissions by 30%

per time unit, shipping lines started to operate in slow steaming mode in the last years.

In slow steaming mode, feeder lines could also decrease waiting idle times of their

3. Feeder Service 42

feeder ships in hub ports for the next voyage. Feeder vessels could operate on different

speeds in different steps of the voyage. A ship usually operates faster speeds in high

demand volume direction and slower speeds in low demand directions (Christiansen et

al. 2007; Windeck 2013).

Stowage planning: Stowage planning decisions are related to the positioning of con-

tainers on vessel’s board. Efficient stowage planning is essential for a good weight bal-

ance and sailing stability of the vessels. Moreover, since the access to the containers is

only possible from the top of the stack, efficient stowage planning reduces the number

of unnecessary shifts in unloading of the delivery containers and berthing time of ves-

sels at feeder ports.

Environmental routing: Environmental routing considers the optimal sailing path

between two ports in route sequence by considering water depths, tides, regulations, and

direction and speed of waves and winds (Windeck 2013).

Empty container repositioning: The trade imbalance in the region results in the emp-

ty container repositioning problem. In order to supply additional empty container re-

quirement of import intense ports, the feeder lines have to transport surplus empty con-

tainers of export intense ports to import ports. Also the route design should consider the

maximum deadweight of the vessels by mixing empty and full containers to increase

utilization of vessels.

4. Literature Review 43

4. Literature Review

The related literature review summarized in five parts in this chapter. While Section 4.1

summarizes the recent related liner shipping network design papers, Section 4.2 over-

views the feeder service management and feeder network design related papers. Section

4.3 gives a basic categorization for vehicle routing problems and summarizes the

VRPSPDTL papers in the literature. Section 4.4 give a compression over the liner ship-

ping related forecasting approaches and Section 4.5 summarizes the studies which han-

dle liner shipping problems under unstable demand environments.

4.1 Liner shipping network design

Maritime liner shipping has become a popular topic of academic research worldwide.

Hence, a huge amount of papers has been published focusing on different planning as-

pects in this area. A large number of papers addresses ship routing and scheduling, cf.

Ronen (1983, 1993), Christiansen et al. (2004; 2013), Kjeldsen (2011) and Meng et al.

(2013) for comprehensive reviews. Additional review papers appeared on container

shipping (Notteboom 2004), fleet size and mix (Pantuso et al. 2013), fleet composition

and routing (Hoff et al. 2010) and liner shipping network design (Ducruet and

Notteboom 2012; Yang et al. 2012). Because of the availability of some detailed litera-

ture reviews, we highlight some recent papers which are considered more related to lin-

er shipping feeder service network design in this chapter. In Table 4.1 an overview of

these papers is given.

In the liner shipping service network design literature, ever since the studies of Rana

and Vickson (1988, 1991), much attention has been paid to routing and scheduling of

the ships in MPC shipping networks. Ting and Tzen (2003) proposed a mathematical

model for service network design of MPC liner shipping line by considering service

time windows. In order to minimize total network cost, the authors developed a dynam-

ic programming model and solved a case study for transatlantic service with 10 ports.

4. Literature Review 44

Table 4.1: Liner shipping service network design related studies

Authors and Years

Problem

Network

Formulation

Object function

Solution Methodology

Feeder

Case Area

Ting and Tzen

(2003)

Fleet deployment, Ship sched-

uling, Ship routing

MPC

Dynamic programming

Min. total cost

Transatlantic

Shintani et al.

(2007)

Ship routing, Empty container

repositioning,

MPC

Cost model formulation

Max total profit

Genetic Algorithm

Southeast Asia

Lei et al. (2008)

Ship scheduling, Collaboration

MPC

Mixed integer linear pro-

gramming

Min. total cost

CPLEX

Randomly generated

Yan et al. (2009)

Ship scheduling

MPC

Integer multiple commodity

network ﬂow

Min. total cost

CPLEX

Taiwan

Lam (2010)

Fleet deployment, Ship sched-

uling, Ship routing

MPC

Cost model formulation

Max total profit

Intelligent system

Wang and Meng

(2011)

Ship scheduling

MPC

Mixed integer linear pro-

gramming

Max total profit

Two staged column gen-

eration

Asia-Europe

Windeck (2013)

Ship routing, ship scheduling,

environmental routing

MPC

Mixed integer programming

Max total profit

Variable Neighborhood

Gulf of Mexico, N.

Atlantic, North Sea

Gelareh and Meng

(2010), Wang et al.

(2011)

Service frequency, Fleet de-

ployment, Ship scheduling,

Chartering

MPC

Mixed integer linear pro-

gramming

Min. total cost

CPLEX

Transpaciﬁc, Transat-

lantic, Asia– Oceania

Europe

Ronen (2011)

Fleet deployment, Bunker

cost, Slow streaming

MPC

Cost model formulation

Min. total cost

Various

Meng and Wang

(2011b)

Fleet deployment

MPC

Mixed integer linear pro-

gramming, Dynamic pro-

gramming

Max total profit

CPLEX

Transpacific, South-

east Asia - Oceania,

intra-Indian Ocean

Agarwal and Ergun

(2008)

Ship scheduling, Ship routing

Interlining

Mixed integer linear pro-

gramming

Max total profit

Column generation,

Benders decomposition

Randomly generated

Alvarez (2009)

Ship routing, Fleet deployment

Interlining

Mixed integer programming

Min. total cost

Column generation,

CPLEX, Tabu Search

Global

Lachner and

Boskamp (2011)

Ship routing, Ship scheduling

Interlining

Cost model formulation

Max total profit

Multi-start local search

heuristics

Asia-Europe

Reinhardt and

Pisinger (2012)

Ship routing, Ship size

Interlining

Mixed integer linear pro-

gramming

Min. total cost

CPLEX

Randomly generated

Wang and Meng

(2012a)

Fleet deployment, Chartering

Interlining

Mixed integer linear pro-

gramming

Min. total cost

CPLEX

Asia–Europe–Oceania

Wang et al. (2013c)

Container path routing

Interlining

Integer linear programming

Min routing cost

CPLEX

Global

4. Literature Review 45

Authors and Years

Problem

Network

Formulation

Object function

Solution Methodology

Feeder

Case Area

Hsu and Hsieh

(2005)

Ship routing, Ship size, Sailing

frequency

MPC, H&S

Multi-Objective Model

Min. shipping costs,

inventory costs

Pareto optimal solutions

Shuttle

Transpaciﬁc

Løfstedt et al.

(2010)

Complex network design

MPC, H&S

Rich integer programming

model

Max total profit

Benchmark sets

Chen and Zhang

(2008)

Ship routing, Ship size

MPC, H&S

Mixed integer linear pro-

gramming

Min. total cost

TSP heuristic, Minimum

location

Shuttle

Asia-Europe, Asia-

North America

Imai et al. (2009)

Ship routing, Ship size, Empty

container repositioning

MPC, H&S

Mixed integer linear pro-

gramming

Min. total cost

LINGO

Shuttle

Asia-Europe, Asia-

North America

Meng and Wang

(2011a)

Empty container repositioning,

Combined MPC& H&S

MPC, H&S

Mixed integer linear pro-

gramming

Min. total cost

CPLEX

Cyclic

Asia–Europe–Oceania

Kjeldsen (2011)

Ship routing, Container rout-

ing

H&S

Mathematical flow model

Max total profit

Dantzig-Wolfe decompo-

sition, column generation

Cyclic

Africa, Europe, North

America

Takano and Arai

(2009)

Hub location and spoke alloca-

tion

H&S

Uncapacitated single-

allocation p-hub median

Min. total cost

Genetic Algorithm

Shuttle

East Asia

Gelareh et al. (2010)

Hub location and spoke alloca-

tion

H&S

Mixed integer programming

Max market share

Lagrangian relaxation

and CPLEX

Shuttle

Randomly generated

Gelareh and Nickel

(2011)

Hub location and spoke alloca-

tion

H&S

Mixed integer programming

Min. total cost

Benders decomposition,

Greedy neighborhood

search, CPLEX

Shuttle

AP dataset

Gelareh and Pisinger

(2011)

Hub location and spoke alloca-

tion, fleet deployment

H&S

Mixed integer linear pro-

gramming

Max total profit

Benders decomposition,

CPLEX

Shuttle

Randomly generated

for North-American

and European

Zacharioudakis et al.

(2011)

F. deployment, Slow stream-

ing, Frequency, Chartering

H&S

Generic cost model

Min. total cost

Genetic Algorithm

Cyclic

Transpaciﬁc & Intra-

Asia

Mulder (2011)

Fleet deployment, Container

path routing, Ship scheduling

H&S

Linear programming model

Max total profit

Genetic Algorithm

Cyclic

Asia-Europe

Hsu and Hsieh

(2007)

Ship routing, Ship size, Sailing

frequency

H&S

Multi-Objective Model

Min. shipping costs

and inventory costs

Pareto optimal solutions

Shuttle

Transpaciﬁc

Yang and Chen

(2010)

Ship routing, Ship size, Sailing

frequency

H&S

Bi-level programming

Min. transportation

cost

Genetic Algorithm

Cyclic

China –West America

Lu and Meng (2011)

Ship routing, Ship size

H&S

Cost model formulation

Min. operation cost.

Tabu Search

Cyclic

East Asia - South Asia

- North Europe

4. Literature Review 46

Their results show that optimization based models could help planners to make better

estimation for voyage fixed costs and freight variable costs in MPC liner shipping net-

work design.

Shintani et al. (2007) developed MPC container shipping network design model by

considering the empty container repositioning. In order to maximize total profit from

the network, the authors proposed a genetic algorithm based solution approach and

solved case studies from Southeast Asia region with 20 ports and 3 ships. Lei et al.

(2008) developed service network design for a MPC liner shipping line by considering

service time windows and different collaboration policies between carries. In order to

minimize total transportation cost, the authors developed mixed integer linear pro-

gramming models for the non-collaborative policy, the slot-sharing policy, and the total-

collaboration policies and solved a number of randomly generated test instances by us-

ing CPLEX. Their results show that the total- collaboration policy between carriers has

great potential to decrease total network cost. Yan et al. (2009) proposed a mathematical

model for short-term fleet scheduling problems in already designed MPC service routes.

In order to minimize total network cost, authors formulated the problem as integer mul-

tiple commodity network ﬂow and solved Lagrangian relaxation with sub-gradient

method by using CPLEX. The proposed model has been implemented to liner shipping

network of a Taiwanese liner shipping provider who serves 11 ports in East Asia.

Lam (2010) presented an integrated approach for designing MPC service routes by

considering time windows and suggested an decision support system for the problem.

The proposed approach selects the calling ports from the regions’ candidate ports ac-

cording to preferences of carriers and offers optimal routes, fleet deployment and

scheduling for the selected ports by minimizing total network cost. Wang and Meng

(2011) developed service network design for MPC liner shipping line by considering

service time windows. In order to maximize total profit from the network, the authors

proposed a two staged column generation based solution approach with numerical ex-

amples from the Asia-Europe liner service. Windeck (2013) proposed a mixed integer

programming model for joint routing and scheduling of MPC shipping lines under envi-

ronmental influences to maximize total network profit. The authors developed a Varia-

ble Neighborhood Search for case studies with up to 33 ports and 6 ships.

A field concerning the deployment of the ships to already designed MPC liner ship-

ping networks is one of the recent study areas. Gelareh and Meng (2010) developed a

4. Literature Review 47

joint service frequency and fleet deployment model to already designed liner shipping

MPC service network by considering service time windows and vessel charter opera-

tions. In order to minimize total network cost, the authors proposed a mixed integer lin-

ear programming model and solved transpaciﬁc service with 7 ports, transatlantic ser-

vice with 6 ports, and Asia–Europe service with 9 ports by using CPLEX. Later, Wang

et al. (2011) reformulate the fleet deployment model of Gelareh and Meng (2010) in

order to increase efficiency of the model. Ronen (2011) developed fleet development

model to already designed liner shipping MPC service networks by considering differ-

ent levels of bunker costs and vessel speeds. In order to find optimal speed for minimiz-

ing network cost, the authors developed a mathematical formulation and solved some

numeric case studies exists in the literature. Their results show that operating under

lower speed has potential to reduce total operating costs but increases transit times.

Meng and Wang (2011b) proposed an integer linear programming model to determine

fleet size, mix and deployment on already designed service network during yearly peri-

ods. The shipment demands between ports are also given and increases yearly 10% dur-

ing the periods.

Another common research area is designing interlining service networks of trunk

shipping lines by considering transshipment operations. Agarwal and Ergun (2008) de-

veloped interlining service network design for liner shipping lines with transshipment

operations by considering simultaneous ship routing and scheduling problem without

considering transshipment cost. In order to maximize total network profit, the authors

proposed a mixed integer programming model and solved the joint problem by using a

greedy heuristic, a column generation based algorithm and a two phase Benders decom-

position based algorithm. The authors solved a number of randomly generated instances

consisting up to 20 ports and 100 ships. Alvarez (2009) developed a global sized inter-

lining service network design by considering simultaneous ship routing and fleet de-

ployment problem. In order to minimize total network cost, the authors proposed a

mixed integer programming model and solved the joint problem by using Tabu Search

heuristics and CPLEX. The authors solved an assumption based case study of global

container trade number with 2.6 million TEUs demand of 120 ports.

Lachner and Boskamp (2011) proposed a joint ship routing and scheduling model

for interlining service network design problem to maximize total network profit. The

authors proposed a multi-start local search heuristic with several variations and solved

4. Literature Review 48

case studies from Asia-Europe shipping network with 58 ports. Reinhardt and Pisinger

(2012) developed interlining service network design for trunk lines of liner shipping

with transshipment operations by considering fleet assignment problems. In order to

minimize total network cost, the authors proposed a mixed integer linear programming

model and solved problem up to 15 ports by using branch-and-cut. Wang and Meng

(2012a) also developed fleet deployment model to already designed liner shipping inter-

lining service network with transshipment operations by considering vessel charter op-

erations. In order to minimize total network cost, authors proposed a mixed integer line-

ar programming model and solved Asia–Europe–Oceania shipping network with 46

ports by using CPLEX. Wang et al. (2013c) proposed an integer linear programming

model for generating optimal container routing paths in already designed network in

order to minimize transportation cost per container while considering the transit time

and maritime cabotage constraints. The authors applied model to a liner shipping net-

work provided by a global liner shipping company with 166 ports from all over the

world, 75 ship routes, and 538 voyage legs.

Just a few papers deal with the economic evaluation of the MPC and H&S network

designs. Hsu and Hsieh (2005) developed a two-objective model for container shipping

lines in order to compare the optimal routing, ship size, and sailing frequency decisions

for MPC and H&S service networks. In order solve the model, the authors used Pareto

optimality concept instead of a complete optimal solution and solved the model for

transpaciﬁc shipping service case study with 5 ports. Their results show that optimal

decision tends to be direct shipping as container flow between origin and destination

ports increases for such a small networks. Chen and Zhang (2008) compared economic

viability of mega-size containership in both MPC and classical H&S network design by

considering transshipment and feeder service costs. In order to minimize total network

cost, authors proposed a mixed integer linear programming model and solved Asia-

Europe network with 17 ports and Asia- North America network with 12 ports by using

travelling salesman heuristics for MPC and minimum location for H&S. Their results

show that Mega containers were competitive in Asia-Europe trade in all scenarios and

competitive for Asia- North America as long as feeder costs are low.

Imai et al. (2009) also compared MPC and classical H&S network service designs

by including the empty container repositioning for both MPC and H&S service net-

works of liner shipping. The authors proposed a number of numerical experiments for

4. Literature Review 49

Asia–Europe and Asia–North America trade lanes and solved the problem with the help

of LINGO solver. Their results show that MPC has total cost advantages in Asia-North

America shipping network and H&S has total cost advantages in Asia-Europe shipping

network. Meng and Wang (2011a) compared the cost-effectiveness of already designed

pure H&S, pure MPC, and combined H&S and MPC service networks by including the

empty container repositioning. The authors formulated a mixed-integer programming

model in order to minimize total operation cost. The authors proposed a case study for

Asia–Europe–Oceania shipping service network consisting of 46 ports and related 24

test instances are solved by using CPLEX. Their results show that combined network

design is more cost-effective than pure H&S and pure MPC service network. Kjeldsen

and Lysgaard (2011) presented a mathematical flow model for joint routing of ships and

cargo in H&S and MPC liner shipping service networks by including various deploy-

ment issues to maximize total network cost. The authors proposed a heuristic is devel-

oped based on Dantzig-Wolfe decomposition and column generation and solved case

studies with various network instances in Africa, Europe, and North America with up to

25 ports and 45 ships by using CP.

A growing number of studies aim to determine optimal configuration of hub loca-

tions and allocation of feeder ports to hub in H&S networks. Baird (2006) presented a

methodology for evaluating and comparing hub ports in Northern Europe. Shuttle feed-

er shipping costs for current hub locations and a newly proposed hub port in the Orkney

Islands are compared. Takano and Arai (2009) considered a hub location problem in

H&S service network design with shuttle feeder service by considering allocation of

feeder ports to hub ports. In order to minimize total network cost, the authors formulat-

ed the problem as an uncapacitated single-allocation p-hub median and solved problem

by using Genetic Algorithm approach with a case study from East Asia containing 14

ports. Gelareh et al. (2010) also considered a hub location problem in H&S service net-

work design under competition between a newcomer liner service provider and an exist-

ing dominating operator. In order to maximize market share of new comer line, the au-

thors proposed a mixed integer programming model and solved the case studies up to 72

ports with the help of Lagrangian relaxation and CPLEX.

Gelareh and Nickel (2011) proposed a formulation for the uncapacitated multiple al-

location hub location problem for H&S liner shipping network design. In order to min-

imize total network cost, the authors developed a mixed integer programming model

4. Literature Review 50

and solved the problem by using a Benders decomposition with CPLEX for small sized

test instances and Benders decomposition with greedy neighborhood heuristic for large

sized test instances. Gelareh and Pisinger (2011) developed a joint hub location and

fleet deployment model for H&S networks with direct feeder services for container

transportation. In order to maximize total network profit, the authors proposed a mixed

integer linear programming model formulation solved a number of randomly generated

test instances for North-America Asia trade by using benders decomposition model and

CPLEX.

Fleet deployment of designed shipping networks is also one of the recent study areas

in H&S liner shipping networks. Zacharioudakis et al. (2011) proposed a fleet deploy-

ment model for already designed H&S service network by considering various service

options. The authors developed a generic cost model methodology to minimize total

network costs by using Genetic Algorithms and solved transpacific shipping trunk ser-

vice network including cyclic intra-Asia feeder routes with 20 ports. The proposed

model decreased the total operation cost of the network almost 36%. Mulder (2011)

developed a combined fleet deployment, ship scheduling and container routing path

problem model in an already designed H&S service network by considering both trunk

and feeder line operations. The author proposed a linear programming model to maxim-

ize total network cost and provided a genetic algorithm based approach to solve case

studies from Asia-Europe shipping network consisting 58 ports. Proposed model im-

proved the total network profit about 40% compared to the reference network.

In the literature, only a few papers have been published which consider H&S sys-

tems with origin-to-destination transportation processes as a whole. Hsu and Hsieh

(2007) presented a two-objective model for H&S service network model in order to

determine the optimal routing, ship size, and sailing frequency by minimizing shipping

costs. In order solve the model, the authors used Pareto optimality concept instead of a

complete optimal solution and solved the model for transpaciﬁc shipping service case

study with 7 ports. Løfstedt et al. (2010) developed a benchmark suite for liner shipping

by including various types network design problems along with a rich integer program-

ming model. The authors also presented a set of benchmark data instances with oﬀset in

real world data.

Yang and Chen (2010) considered a joint routing, ship size, sailing frequency prob-

lem for a modern H&S service network design with cyclic feeder service by considering

4. Literature Review 51

allocation of feeder ports to hub ports. In order to minimize total operating cost per

voyage, the authors proposed a bi-level programming model formulation and solved a

case study China-West America with 14 ports by using Genetic Algorithm. The authors

used trunk lines for high demanded main ports and hub ports, and feeder services for

low demanded small ports in order to avoid too much port calling of trunk lines. Lu and

Meng (2011) proposed a solution for simultaneously routing problem and ship size of

trunk and feeder lines in a modern H&S service network by considering allocation of

feeder ports to hub ports. In order to minimize total operation cost per voyage, the au-

thors mathematically formulated the problem and proposed a Tabu Search based solu-

tion approach to solve East Asia - South Asia - North Europe liner service network con-

sisting 11 hub ports and main ports and 51 feeder ports. The authors did not include

capital costs of the ships and the necessary number of ships per voyage.

4.2 Feeder service

Nowadays feeder services play an irreplaceable role in global shipping networks. The

related literature primarily addresses regional conditions and prospects in the develop-

ment of feeder services. For instance, Waters (1973) searched adaptability of full coastal

container shipping services by comparing different types of ships to feed main line ports

in New Zealand. The author suggested using of only small type containerships due to

lack of equipment of faraway small ports. Robinson (1998) discussed the development

of Asian ports between 1970 and 2000. The authors reported that during the years, op-

eration and organization of direct, feeder and trunk service networks always changed

parallel to development of the region. The author argued the importance of feeder net-

work will continue to grow over a long period of time. Frankel (2002) analyzed the

feeder service structure Caribbean ports and compared economics of shuttle, cyclic,

pendulum feeder service operations with direct trunk service for three regional ports.

The authors suggested use direct trunk services or shuttle services to region ports under

low level of bunker prices. Evers and Feijter (2004) discussed feeder ship service opera-

tions in the Rotterdam port.

Ridolfi (1999) examined problems and prospects of global and regional shipping

lines in the Mediterranean region. The author argued that with the increasing volume of

transshipment trafﬁc services, feeder service networks will continue to stimulate the

regional transportation role at many more terminals around the region. In addition,

4. Literature Review 52

Francesetti and Foschi (2002), Goulielmos and Pardali (2002), Foschi (2003), Ham and

Autekie (2005) and Jadrijević and Tomašević (2011) discussed prospects and problems

of Mediterranean Sea feeder services and ports.

Kolar et al. (2000) analyzed necessities of feeder service from a small regional port

Rijeka (Croatia) to a hub port Gioia Tauro (Malta) in Mid- Mediterranean region. 5

years later after starting of the feeder service between Rijeka and Gioia Tauro, Hlača

and Babić (2006) evaluated the effect of the feeder service to region and the current

conditions and the future prospects of the feeder service. Buksa and Kos (2005) and

Rudic and Hlaca (2005) analyzed the structure of feeder containership services in the

Adriatic ports. In addition to these studies, few studies concentrated on CCS line net-

works. Luo and Grigalunas (2003) simulated demand of CCS line networks for US

coastal ports. Ran et al (2008) analyzed characteristics of Chinese CCS market.

Further analyses on determinants of container liner shipping are recently investigat-

ed by couple of researchers. Chang et al. (2008) compared port selection behaviors of

trunk liner and feeder line providers. The authors argue that local cargo volume, termi-

nal handling charge, land connection, service reliability, berth availability, transship-

ment volume and cargo proﬁtability are the main determining factor for feeder service

providers. Ng and Kee (2008) evaluated optimal containership size of shuttle feeder

services by using economic and simulation models in Southeast Asia from the perspec-

tive of carriers. Styhre (2010) analyzed the factors affecting utilization of feeder service

vessels. The author argue that market, customer, port and surrounding factors are exter-

nal factors and management and vessel factors are internal factors in the utilization per-

formance of feeder lines. Recently, Buksa and Zec (2011) and Buksa and Buksa (2011)

analyzed risk and quality issues of coastal container liner shipping lines.

Recently, Varbanova (2011a, b) analyzed transportation conditions and efficiency

factors of feeder lines in the Black Sea region. The author also summarized the opera-

tional planning problems of container feeder lines which are related to schedule integri-

ty between trunk lines, shipping capacity utilization, service speed optimization under

high bunker prices, recent regulations for control of air pollution from ships, shipping

policy impulsions of the European Union, structure and capacity of the container termi-

nals, and increased dwell time of containers at the terminals.

4. Literature Review 53

Nowadays feeder services play an irreplaceable role in global shipping networks.

Just a few papers deal with real feeder service network settings and consider existing or

projected hub locations. Bendall and Stent (2001) developed a mixed integer linear pro-

gramming model for deployment and scheduling of ships in an already designed feeder

service network. The proposed model, firstly, determines the number of voyages to be

commenced on each route in order to maximize total network profit by sometimes de-

clining some low profit and secondly, schedules ships in the determined voyages. The

authors solved a case study from South East Asia with six ports and eight voyages and a

couple of homogenous fast feeder containerships. Mourao et al. (2001) proposed an

integer linear programming model for assignment of a number of ships to an already

designed network containing one feeder route and two trunk routes by considering de-

livery containers and inventory costs. Catalani (2009) proposed a cost-minimization

based expert system model for sequencing and scheduling feeder ports for just one

feeder service route within the Mediterranean area with one hub ports and four feeder

ports by considering time windows and both pick-up and delivery containers.

Other papers put a stronger focus on mathematical modeling and related solution

methodologies with unique feeder service configurations. Chou et al. (2003) presented

sea freight inventory-routing problem by considering both direct and feeder service to

distribute demands of regional ports from a supply port by using a fleet of homogenous

ships. Direct shipping modeled as traveling repairman problem and feeder service mod-

eled by using mixed integer problem formulation to minimize the average in-transit in-

ventory cost. The proposed models are solved by using CPLEX and Multi-start Tabu

Search algorithm and compared in a case study from South East Asia with 10 ports. One

example is the paper by Baird (2006) who presents a methodology for evaluating and

comparing hub ports in Northern Europe. Based on mainline ship deviations and direct

feeder shipping costs increased transshipment capacities of ports are given.

Sigurt et al. (2005) and Andersen (2010) presented a regional feeder service network

by allowing direct and indirect delivery between feeder ports, recurring visits and con-

sidering time windows. The authors proposed different mathematical modeling formula-

tions and solution approaches based on decomposition of the problem into two sub-

problems dealing with master problem and freight routing. Firstly, the authors initialize

a number alternative route by using one hub port and determine optimal service network

design by deploying ships to these routes. Then, allocates a number of origins to desti-

4. Literature Review 54

nation freights to these routes in order minimize total freight routing cost in the net-

work.

In the academic literature, only a few studies treated the feeder service problem as a

variant of vehicle routing problem. Fagerholt (1999; 2004) considered feeder shipping

problem in a special network where all pick-up cargo are collected from a set of feeder

ports to a single hub port by using various size ships with same speed. The problem is

solved by first initializing all feasible single ship routes according to biggest sized ship

by implementing time limits, and then allocating optimal ship types to these routes.

Contrary to classic vehicle routing problem approach, ships could operate more than

one route (multi-trip) in allowed time limit (VRPMTTL). The author solved instances

with up to 40 ports by using up to 19 ships in 23 routes within a couple of seconds.

Jin et al. (2005) evaluated feeder containership routing problem as a VRP with

pickup and delivery with time windows formulation. The authors proposed a mixed

integer programming model for minimizing total weighted cost, which is a weighed sum

of the total travel times, the number of ships used, total waiting time, and total tardiness

time in order to serve feeder ports from one hub port with a fleet of homogenous ships

under port time windows constraints. The proposed model is solved for VRP test in-

stances up to 100 ports by using Variable Neighborhood Search and Tabu Search

aproaches. Later, Sun and Li (2006) also handled same problem by using immune Ge-

netic Algorithm approach and improved the solutions of Jin et al. (2005)’s the test in-

stances.

Sambracos et al. (2004) present a case study of dispatching small containers via coastal

freight liners from a hub port to 12 Greek island ports. In this study total operating costs

including fuel consumption and port charges are minimized assuming a homogeneous

fleet and given container shipment demand. The authors formulate a linear program-

ming model which is solved by use of a list-based threshold acceptance heuristic. Later

Karlaftis et al. (2009) generalized this container dispatching problem by minimizing

total travel distances with simultaneous container pick-up and delivery operations and

time deadlines. To solve the problem they propose a mixed-integer programming for-

mulation and a Genetic Algorithm assuming soft time limits, i.e. tolerating violations of

4. Literature Review 55

Table 4.2: Overview of feeder service network design related studies

Authors and

Years

Problem

Freight

Fleet

Formulation

Object function

Solution Methodology

Feeder

service

Case Area

Bendall and

Stent (2001)

Fleet deployment, ship

scheduling, Recurring

visit

Container

Homogenous

Mixed integer pro-

gramming

Max. total network

profit

Branch and Bound

Shuttle,

Cyclic

South East

Asia

Mourao et al.

(2001)

Fleet size, inventory

Container

Homogenous

Integer linear pro-

gramming

Minimize total annual

trade cost

Excel solver

Cyclic

Portugal

Catalani (2009)

Port sequencing, Ship

scheduling

Container

Homogenous

Mathematical cost

modeling

Min. operating cost

Expert System

Cyclic

Aegean Sea

Chou et al.

(2003)

Ship inventory-routing

Ores

Homogenous

Mixed integer pro-

gramming

Min. average in-

transit inventory cost

CPLEX, Multi-start

tabu search

Cyclic

South East

Asia

Sigurt et al.

(2005)

Master problem, Ship

scheduling, Freight

routing

Cargo

Heterogeneous

Mixed integer linear

programming

Min total freight rout-

ing cost

Linear programming

relaxation, Heuristic

branch-and-price

algorithm

Cyclic,

recurring

visit

Randomly

generated

Andersen (2010)

Master problem, Ship

scheduling, Freight

routing

Container

Heterogeneous

Mixed integer linear

programming

Min total freight rout-

ing cost

Linear programming

relaxation, CPLEX

Cyclic,

recurring

visit

Randomly

generated

Fagerholt (1999;

2004)

Ship routing, fleet size

and mix (VRPMTTL)

Container

Heterogeneous

Integer programming

Min total transporta-

tion cost

Set partitioning,

CPLEX

Shuttle,

Cyclic

Randomly

generated

Jin et al. (2005)

Ship routing

(VRPPDTW)

Container

Homogenous

Mixed integer pro-

gramming

Min. total weighted

cost

Variable neighbor-

hood search, Tabu

Cyclic

VRP instanc-

Sun and Li

(2006)

Ship routing

(VRPPDTW)

Container

Homogenous

Mixed integer pro-

gramming

Min. total weighted

cost

Immune genetic algo-

rithm

Cyclic

VRP instanc-

Sambracos et al.

(2004)

Ship routing (CVRP)

Container

Homogenous

Plain linear pro-

gramming problem

Min operation cost

List-based threshold

Acceptance heuristic

Cyclic

Aegean Sea

Karlaftis et al.

(2009)

Ship routing

(VRPSPDTL)

Container

Homogenous

Mixed integer linear

programming

Min total travel time

Genetic algorithm

Cyclic

Aegean Sea

4. Literature Review 56

certain constraints. As a case study they deal with a feeder network from the Aegean

Sea with 26 ports including one hub port. Nevertheless, both studies did not consider

heterogeneous ship types and the specific costs for operating the fleet. See Section 5.2

for mathematical model details of containership routing problem of Karlaftis et al.

(2009) and Section 7.2 for implementation of the problem with developed hybrid solu-

tion heuristic called Adaptive Neighborhood Search (Section 6.1).

However, so far none of the feeder service network design studies has considered

detailed variable and fixed cost structures for a heterogeneous vessel fleet over a com-

plete sailing horizon. In this study, we propose a mixed-integer linear programming

model to simultaneously determine the fleet size and mix, fleet deployment, ship rout-

ing and ship scheduling in feeder service network design by minimizing total network

costs in a sailing season (see Section 5.3). A hybrid solution heuristic called Adaptive

Neighborhood Search (see Section 6.1) is proposed to solve the joint problem using a

case study from Black Sea region (See Section 7.3).

4.3 Vehicle routing problem

Since basic specifications of feeder service network design problem fundamentally fits

to the vehicle routing problem with simultaneous pickup and delivery with time limit

(VRPSPDTL ) (Polat et al. 2012), this section provides a brief literature over VRP vari-

ations and VRPSPDTL applications.

The Vehicle Routing Problem (VRP) refers to serving a set of clients from a central

depot with a homogeneous ﬂeet of capacitated vehicles. This problem aims to determine

a set of vehicle routes starting and finishing at the central depot which serves all clients

just once, thereby minimizing the total distance travelled. A variation of the VRP is the

Vehicle Routing Problem with Pickup and Delivery (VRPPD) where the vehicles serve

both delivery and pick up operations at client locations. The VRPPD can be basically

categorized into three classes (Nagy and Salhi 2005):

– VRP with Backhauls (VRPB): the vehicles first serve delivery operations, next

pick up operations at clients.

– Mixed VRPB (MVRPB): the vehicles serve delivery or pick up operations to cli-

ents in any sequence.

4. Literature Review 57

– VRP with Simultaneous Pickup and Delivery (VRPSPD): the vehicles simultane-

ously serve delivery and pick up operations to clients.

Please see Berbeglia et al. (2007) and Parragh et al. (2008) for extended variants of

Pickup and Delivery Problems (PDP). In the literature, the VRPSPD was first proposed

by Min (1989). In this problem, the existing load of the vehicle has to be checked at

each client to ensure that the vehicle capacity is not violated. The VRPSPD can also be

categorized into three classes:

– VRPSPD with Maximum Distance Length (MLVRPSPD): the vehicles have a

maximum voyage distance constraint for returning to the central depot.

– VRPSPD with Time Windows (VRPSPDTW): the vehicles have to start their ser-

vice with the clients between a given earliest and latest time.

– VRPSPD with Time Limit (VRPSPDTL): the vehicles have to return to the central

depot before a time deadline.

In this study, we consider the VRPSPDTL which was first proposed by Salhi and

Nagy (1999). The authors impose service times for the clients and a maximum total

duration (travel + service time) restriction for the vehicles in the VRPSPD.

Dethloff (2001) defined the VRPSPD as an NP-hard combinatorial optimization

problem, meaning that practical large-scale problem instances are hard to solve through

exact solution methodologies within acceptable computational times. In the

VRPSPDTL the objective and constraints are the same as in the VRPSPD, except for

the service time limit of vehicles. This makes the problem more complicated due to the

difficulty in controlling of the voyage duration of the vehicle in addition to the service

time of the clients along the route. As a result, this problem can be described as NP-

hard, as well. In the literature, the interest was therefore focused on heuristic or

metaheuristic solution approaches.

Since 1989, many heuristic and meta-heuristic solution approaches for VRPSPD

benchmark problems have been proposed. See Zachariadis et al. (2009), Subramanian et

al. (2010), Zachariadis and Kiranoudis (2011) and Goksal et al. (2012) for recent studies

on the VRPSPD. See Tang and Galvao (2006), Zhang et al. (2008) and Fard and Akbari

4. Literature Review 58

(2013) for recent studies on the MLVRPSPD and Mingyong and Erbao (2010) for the

VRPSPDTW.

However, just a few studies considered benchmark problems under time limit re-

strictions. Some authors proposed heuristic and metaheuristic implementations for the

VRPSPD as well as VRPSPDTL.

Salhi and Nagy (1999) presented single and multi-depot VRPB benchmark problems

including VRPB, VRPPD, VRPSPD and VRPSPDTL. The authors manipulated 7 origi-

nal single depot VRP benchmark problem instances of Christoﬁdes et al. (1979) by im-

posing a maximum time restriction for the vehicles, giving a predefined service time,

and splitting the original demand between pickup and delivery loads in order to create

VRPSPDTL test instances. The remaining 7 instances were obtained by switching these

pickup and delivery loads. The authors proposed a Cluster Insertion Heuristic (CIH) to

solve problem in order to minimize the total travelled distance covered by a fleet of ve-

hicles.

Later from Salhi and Nagy (1999), a number of studies improved best known solu-

tions day by day with various heuristic and metaheuristic approaches. Detholff (2001)

proposed an Insertion Based Heuristic (IBH) for solution of VRPSPDTL instances. The

authors tested total distance, residual capacity, radial surcharge, and combination of

residual capacity and radial surcharge insertion criterions within proposed IBH. They

got their best results with combination insertion criterion for VRPSPDTL test instances.

Nagy and Salhi (2005) developed a number of heuristics by integrating various neigh-

borhood structures. The authors catch their best results in VRPSPDTL by using alternat-

ing integration of these heuristics called as Alternating Heuristic (ALT).

Ropke and Pisinger (2006) used a Large Neighborhood Search (LNS) approach by

using a number of removal and insertion heuristics with and without learning layers to

solve test instances. They got their best results by using six removal heuristics with

learning layers. Montane and Galvao (2006) developed a Tabu Search (TS) algorithm to

solve VRP and TSP instances including VRPSPDTL. The developed algorithm uses

grouping and routing heuristics to construct an initial solution for TS and three types of

inter-route neighborhoods (the relocation, interchange and crossover) and an intra-route

neighborhood (2-opt) with intensiﬁcation and diversiﬁcation search strategies to im-

prove solutions. Wassan et al. (2008) proposed a Reactive Tabu Search (RTS) algorithm

4. Literature Review 59

which constructs an initial solution with forward and backward-sweep methods and

improves initial solution with three intra-route and one inter-route neighborhoods by

using a dynamic tabu list size.

Gajpal and Abad (2009) presented an Ant Colony System (ACS) approach to solve

problem instances. The presented approach constructs an initial solution using the near-

est neighborhood heuristic, generates routes with savings, and improves the solutions

with the customer insertion/interchange multi-route scheme and the sub-path exchange

multi-route scheme local search heuristics. Ai and Kachitvichyanukul (2009) developed

an Particle Swarm Optimization (PSO) algorithm with multiple social learning struc-

tures by including the cheapest insertion heuristic and 2-opt methods to construct routes.

Catay (2010) Saving Based Ant Algorithm (SBAA) equipped with a new saving-

based visibility function to construct routes and pheromone update procedure to im-

prove solutions within ant algorithm. And recently, Jun and Kim (2012) offered an per-

turbation based solution which construct an initial solution with sweep method, im-

proves the solution with a series of inter- and intra-route neighborhood structures and

perturbs best solution destroy and repair based heuristic when the improvement stuck at

a local optima. The developed approach called as Nearest Sweep with Perturbation

(NSP). All these studies handled VRPSPDTL within VRPSPD problem instances.

Subramanian and Cabral (2008) presented the first investigation that deals with the

pure VRPSPDTL considering the CMT 6-7-8-9-10-13-14 X&Y benchmark problems of

Salhi and Nagy (1999). The authors proposed an Iterated Local Search (ILS) procedure

in order to solve this problem. Their approach constructs an initial solution with greedy

method, improves the solution with variable neighborhood descent method by using six

neighborhood structures and perturbs the best solution with double-bridge perturbation

function when it is necessary.

In all VRPSPDTL benchmark problems of Salhi and Nagy (1999), service time for

all clients are considered the same within the instances. In the corresponding literature,

Salhi and Nagy (1999), Detholff (2001), Nagy and Salhi (2005), Ropke and Pisinger

(2006), Gajpal and Abad (2009), Ai and Kachitvichyanukul (2009), Catay (2010) and

Jun and Kim (2012) solved these instances by including service time, Montane and

Galvao (2006) determined solutions by excluding service time and Wassan et al. (2008)

4. Literature Review 60

and Subramanian and Cabral (2008) considered both situations. Please see Table 4.3 for

brief overview of VRPSPDTL related studies.

In this study, we propose a mixed-integer linear programming model to solve

VRPSPDTL (see Section 5.1) and a hybrid solution heuristic called Adaptive Neigh-

borhood Search (see Section 6.1) is used to solve the benchmark instances, with and

without service time, from Salhi and Nagy (1999) (See Section 7.1)

Table 4.3: Overview of VRPSPDTL related studies

Author (Year)

Solution methods

Abbrev-

iations

With service

time

Without ser-

vice time

Salhi and Nagy (1999)

Cluster Insertion Heuristics

CIH



Detholff (2001)

Insertion Based Heuristics

IBH



Nagy and Salhi (2005)

Alternating Heuristic Algo-

rithms

ALT



Ropke and Pisinger (2006)

Large Neighborhood Search

LNS



Montane and Galvao (2006)

Tabu Search



Wassan et al. (2008)

Reactive Tabu Search

RTS



Subramanian and Cabral

(2008)

Iterated Local Search

ILS



Gajpal and Abad (2009)

Ant Colony System

ACS



Ai and Kachitvichyanukul

(2009)

Particle Swarm Optimization

PSO



Catay (2010)

Saving Based Ant Algorithm

SBAA



Jun and Kim (2012)

Nearest Sweep with Pertur-

bation

NSP



4.4 Container throughput estimation

In the literature, numerous studies have been undertaken on the network design of ship-

ping lines. In these studies authors presented alternative solution models for container

shipping lines in order to determine the optimal routing, ship size, and sailing frequency

by minimizing shipping costs under fixed demand pattern without considering seasonal

demand fluctuations which does not reflect the reality of container shipping network

design.

Since the mid-1950s, forecasting accurate container throughput demands of ports is

one of the major dream of all port economists (Goulielmos and Kaselimi 2011). Be-

cause of the fact that accurately predicting the conditions effect on liner shipping is al-

most impossible, making reliably forecasts ended with certainty is nearly impossible.

But that does not mean forecasting is pointless. The aim of forecasting is not to estimate

4. Literature Review 61

Table 4.4: Container throughput forecasting related studies

Authors (Years)

Proposed Methodology

Comparison Methodologies

Case Area

Walter and Younger (1988)

Iterative Nonlinear Programming

New design

de Gooijer and Klein (1989)

One Vector Autoregressive moving average

One-variable Auto Regression Integrated Moving Average

Antwerp

Zohil and Prijon (1999)

Ordinary least squares regression

Mediterranean

ports

Fung (2001)

Vector Error Correction Model with Structural Identification

Hong Kong

Seabrooke et al. (2003)

Ordinary least squares regression

Hong Kong

Mostafa (2004)

Multilayer Perception Neural Network

Auto Regression Integrated Moving Average

Suez Canal

Lam et al. (2004)

Multilayer Perception Neural Network

Linear Multiple Regression

Hong Kong

Hui et al. (2004)

Error Correction Model Approach

Hong Kong

Guo et al. (2005)

The grey Verhulst model

Grey Model (1,1)

Liu et al. (2007)

Grey Prediction Model and Cubic Polynomial Curve Prediction

Model mixed by the Radial Basis Function Neural Network

Radial Basis Function Neural Network With Grey Prediction Model, Radial

Basis Function Neural Network With Cubic Polynomial Curve Prediction Model

Shanghai

Mak and Yang (2007)

Approximate Least Squares Support Vector Machine

Support Vector Machine, Least Squares Support Vector Machine, Radial Basis

Function Neural Network

Hong Kong

Hwang et al. (2007)

Neuro-Fuzzy Group Method Data Handling Type Neural Networks

Conventional Multilayered Group Method Data Handling Type Neural Net-

works

Busan

Schulze and Prinz (2009)

Seasonal Auto-Regressive Integrated Moving Average

Holt–Winters Exponential Smoothing

Germany

Peng and Chu (2009)

The classical decomposition model,

The trigonometric regression model, The regression model with seasonal dum-

my variables, The grey model, The hybrid grey model, The SARIMA model

Taiwan

Gosasang et al. (2011)

Multilayer Perception Neural Network

Linear Regression

Bangkok

Sun (2010)

Conditional Expectation with Probability Distribution

Shandong

Chen and Chen (2010)

Genetic Programming

X-11 Decomposition Approach, Seasonal Auto Regression Integrated Moving

Average

Taiwan

Wu and Pan (2010)

Support Vector Machine with Game Theory

Jiujiang

Li and Xu (2011)

Prediction Based on Optimal Combined Forecasting Model

Cubic exponential smoothing, GM (1,1), Multiple regression analysis

Shanghai

Goulielmos and Kaselimi

(2011)

The Non-Linear Radial Basis Functions

Piraeus

Zhang and Cui (2011)

Elman neural network

Shanghai

Polat et al. (2011)

Monte Carlo Simulation with Holt–Winters Exponential Smoothing

Turkey

Xiao et al. (2012)

Feed forward neural network is developed based on the improved

particle swarm optimization with adaptive genetic operator

Tianjin

4. Literature Review 62

accurately, it try to help decision-makers to understand the future by reducing uncertain-

ty by exploring the current information. Therefore, forecasting container throughputs of

ports is playing a critical role in the planning decisions of liner shipping lines.

Since liner shipping involves considerable capital investments and huge daily oper-

ating costs, the appropriate liner shipping feeder service network design will affect the

development of the feeder shipping lines. In practice, forecasting container throughput

demands of ports is anywise important for all planning level decisions of feeder service

which are defined in Section 3.7. Table 4.4 summarizes some related studies on con-

tainer throughput forecasting and highlights the methodologies and case studies used in

the related papers.

All these works produced good results under low uncertainty conditions. However,

the 2008 crisis showed that deterministic forecasts may be prone to failure in the long

term (Pallis and de Langen 2010). More realistic stochastic forecasting methods, which

could face uncertainty are not preferred in many practical applications, because of their

complexity and high statistical data requirement (Khashei et al. 2009). Recently, Wang

et al. (2013a) proposed a linear programming model to estimate capacity utilization of

an already sequenced liner ship route with a bounded polyhedral container shipment

demand pattern, but without considering seasonality.

On the other hand, simulation could be used to assist with constructing a forecasting

frame by using deterministic forecasting methods that only need a limited amount of

data. Therefore, in this study, a simulation and artificial neural networks based forecast-

ing framework is developed in order to analyze the impact of seasonal demand fluctua-

tion on the liner shipping feeder service network design (Section 6.2).

4.5 Liner shipping under unstable demand environments

The above literature review clearly shows that container demand uncertainty and sea-

sonality are not well addressed by the liner shipping network design studies. Therefore,

it is important to integrate them into the liner shipping network design problem in view

of their importance in reﬂecting the reality. Liner shipping providers have to deal with

some uncertain and seasonal factors like the real transportation time between two ports,

demand and supply patterns, necessary number and sizes of ships, and the available

capacity of vessels for repositioning empty containers, etc. (see Section 3.6). For in-

4. Literature Review 63

stance, Chuang et al. (2010) developed a Fuzzy Genetic Algorithm approach to define

sequence of ports for a containership in a region. The authors solved a case study with 5

ports to maximize the total profit from the ports without considering ship capacity by

allowing declining low profit ports. The proposed fuzzy sets provide demand fluctua-

tion, but not seasonality.

Since the beginning of containerization, empty container repositioning has been an

on-going issue in the maritime transportation industry. There are several studies that

take into account the uncertain nature of empty container demand parameters in already

designed liner shipping networks. In an early work, Cheung and Chen (1998) proposed

a two-stage stochastic model by considering the uncertainties in container demands and

vessel capacity and their impact on empty container repositioning. Leung and Wu

(2004) developed a multi-scenario time-extended optimization model with stochastic

demands to dispatch empty containers from the Middle East to various export ports in

the Far East region and reposition surplus empty containers from any port to shortage

ports. Feng and Chang (2008) addressed the empty container reposition planning model

by plainly considering safety stock management. The authors considered the model as a

two-stage problem. The first stage was used to estimate the empty container stock at

each port and the second stage reflected the empty container reposition problem in an

already designed liner shipping network of a Taiwanese liner shipping provider.

Song and Dong (2008) considered the empty container repositioning problem in a

dynamic and stochastic situation by minimizing the expected total costs consisting of

inventory holding costs, demand lost-sale costs, lifting-on and lifting-off charges, and

container transportation costs in an already designed network. Performance of a non-

repositioning policy and three other heuristic policies were compared by using a simula-

tion model. Later, Dong and Song (2009) evaluated container fleet sizing problems in

already designed MPC route designs with empty container repositioning under uncer-

tain and imbalanced customer demands which are reflected with uniform and normal

distributions. In order to minimize total network costs, authors formulated a mathemati-

cal model for the problem, proposed a genetic algorithm approach to solve the problem.

They solved problems with a transpacific shipping service which contains 3 ports and 4

vessels with a 22 month period and a Europe-Asia shipping service which contains 10

ports and 8 vessels for a 20 month period. Wang and Tang (2010) proposed a chance-

constrained programming model to maximize the profit of a shipping company under

4. Literature Review 64

uncertain heavy and empty container demands. The authors converted the chance-

constrained programming model to an integer programming model and solved case

studies by using Lingo solver.

Long et al. (2012) formulated a two-stage stochastic programming model with ran-

dom demand, supply, ship weight capacity, and ship space capacity in order to include

uncertainties in the operations model by minimizing the expected operational costs for

empty container repositioning. The authors applied sample average approximation

method to approximate the expected cost function with heuristic algorithms based on

the progressive hedging strategy. Francesco et al. (2009) handled an empty container

repositioning problem by considering demand fluctuations. The authors proposed a

time-extended multi-scenario stochastic optimization model where historical data were

useless for decision-making processes. Later, Francesco et al. (2013) generalized the

empty container repositioning problem of Francesco et al. (2009) by considering possi-

ble port disruptions in an already designed service network.

Some recent studies deal with scheduling containerships in already designed liner ship-

ping networks by considering the uncertain nature of ports. Qi and Song (2012) consid-

er the problem of designing an optimal vessel schedule in an already designed liner

shipping route to minimize the total expected fuel consumption and emissions by con-

sidering uncertain port times and frequency requirements. The authors mathematically

formulated and solved a trans-Pacific container shipping service case study with eight

ports and nine port-of-calls with simulation-based stochastic approximation methods.

Wang and Meng (2012b) considered problems with liner ship schedule designs which

aim to determine the arrival time of all ships at ports by considering the sailing speed

function on each route and uncertainties at sea and port. The authors proposed a mixed-

integer non-linear stochastic programming model to minimize the ship costs and ex-

pected bunker costs. A case study of an already designed Asia-Europe-Oceania shipping

network with 46 ports, 11 ship routes and three types of ships was solved with the mod-

el. Wang and Meng (2012c) considered robust scheduling for liner services by including

uncertain wait time due to port congestion and uncertain container handling time. This

problem is formulated as a mixed-integer nonlinear stochastic programming model

which recovers disruptions in the schedule recovery with fast steaming by keeping pre-

determined port sequence. The authors proposed a hybrid solution algorithm which in-

tegrates a sample average approximation method, linearization techniques, and decom-

4. Literature Review 65

position scheme and solved a case study from an Asia-Europe-Oceania shipping ser-

vice.

The operating efficiency of shipping networks also depends on appropriate slot allo-

cation of containerships. Determination of slot allocations to ports under uncertain de-

mand environment is another recent research area in liner shipping design. Lu et al.

(2010) proposed an integer programming model for slot allocation planning problems of

shipping lines with homogenous ships for satisfying seasonal demands. The authors

solved a real case study from Eastern Asia for a liner shipping service network with 12

ports, 16 sailing legs and deployed 4 containerships with 1445 TEUs capacity. The pro-

posed model works under already forecasted seasonal demands for an already designed

network but does not account for demand fluctuations. Zeng et al. (2010) proposed a

deterministic model to optimize the resource allocation for container lines considering

ship size, container deployment, and slot allocation for shipping lines based on the equi-

librium principle. The authors then converted the deterministic model to a robust opti-

mization model which simultaneously considers demand uncertainty, model robustness,

and risk preference of the decision maker. Zurheide and Fischer (2011, 2012) developed

a simulation model for accepting or rejecting decisions of container bookings by includ-

ing a quantitative slot allocation model with customer segmentations, the service net-

work structure, and transshipment possibilities to maximize net profit of the provider

under different demand scenarios, networks, and input settings.

Containership fleet deployment is a key issue in the liner shipping industry. A num-

ber of recent studies considered fleet deployment in already designed service networks

by considering uncertain demand patterns. Ng and Kee (2008) evaluated optimal con-

tainership size of shuttle feeder services by using economic and simulation models in

Southeast Asia from the perspective of carriers. The authors pointed out that none of the

forecasted container demands of different ports which did not include seasonality and

demand fluctuation were unable to fulfill the simulated optimal ship sizes according to

the opinions of interviewees. Meng and Wang (2010) and Meng et al. (2011) proposed a

chance constrained programming model for short-term liner ship fleet planning to min-

imize total network cost. The proposed model aims to determine fleet size, mix, de-

ployment and frequency on predetermined routes by considering cruising speed. The

authors converted the chance constrained programming model to a mixed integer liner

programming model and solved case studies from an Asia-Europe-Oceania shipping

4. Literature Review 66

service by using CPLEX. The shipment demands between ports are fluctuated by using

normal distribution. Later, Meng et al. (2012) considered the same problem by taking

into account container transshipment operations. The authors proposed a two-stage sto-

chastic integer programming model and a solution algorithm by integrating the sample

average approximation with a dual decomposition and Lagrangian relaxation approach.

Recently, Wang et al. (2012) developed a robust optimization model for the related

problem to maximize total profit under different container demand scenarios. Dong and

Song (2012) evaluated containership fleet sizing problem by simultaneously considering

uncertain customer demands and stochastic inland transport times in an already de-

signed service network. The authors proposed a mathematical formulation for the prob-

lem and solved the case studies from trans-Pacific and Europe–Asia services with the

help of simulation-optimization approaches.

In the dynamic liner shipping literature, considerable attention has been given to se-

quencing of ports, repositioning empty containers, determination of slot allocations,

scheduling of ships, and deploying fleets in already designed service networks by using

demand fluctuations. In addition to these studies, several papers considered these prob-

lems under seasonal demand patterns.

Chen and Zeng (2010) proposed a mixed integer non-linear programming model to

maximize the average unit ship-slot profit with a homogenous ship fleet under seasonal-

ly changing demand and freight rates. The proposed model selects cyclic port sequence

from a number of candidate ports by declining low profit ports and allocates slots to

selected ports. The authors solved a case study from Far East Asia with 10 candidate

ports under average annual and bi-monthly seasonal demand and freight rates by using a

developed bi-level genetic algorithm approach and compared results. Their results show

that designing slot allocations under changing demand and freight rates could increase

maximal total profit 1.41 times and could decrease necessary average slot capacity 0.31

times. Please note that the authors used fixed ship size for the whole year in both sea-

sonal and annual demand patterns, but allocated slot capacities to ports change depend-

ing on seasons.

Meng and Wang (2012) considered fleet deployment and container routing problem

in an already designed liner shipping network with transshipment operations by includ-

ing week dependent origin to destination container demands and maximum allowed

transit time durations. The authors firstly created all possible origin to destination paths

4. Literature Review 67

by using the space–time network approach subject to the transit time constraints in pri-

ori dependent routes. Then by using a mixed integer liner programming model, the op-

timal ships were assigned to ship routes and containers were assigned to paths by con-

sidering week dependent demands in order to minimize total cost while fulfilling the

containership demand. Relaxation models were provided in order to solve the case stud-

ies from an Asia-Europe-Oceania shipping network consisting of 46 ports and 12 routes

with 3 candidate ship types for a 26 week planning horizon. Please note that the authors

deployed ships using fixed ship size for the whole year under weekly demand pattern,

but allocated containers of different origin to destination paths according to seasons.

As is mentioned by Meng and Wang (2012), existing studies in the liner shipping

literature designed liner shipping service networks under fixed demand patterns without

considering seasonal demand fluctuations which does not reflect the reality of liner

shipping network design. On the other hand, the studies considering demand fluctua-

tions and seasonality evaluate specific problems such as slot allocation, empty container

repositioning, ship scheduling, and fleet deployment under already designed service

networks. In this study, we propose a mixed-integer linear programming model to sim-

ultaneously determine the fleet size and mix, fleet deployment, ship routing and ship

scheduling in H&S service network design by minimizing total network costs under

seasonal demand fluctuations in a sailing season (Section 5.4). A simulation-

optimization based solution framework which contains a hybrid solution heuristic called

Adaptive Neighborhood Search (Section 6.1) and a simulation and artificial neural net-

works based forecasting model (Section 6.2) is proposed to solve the joint problem by

using a liner shipping feeder service case study from the Black Sea region (Section 7.4).

Notable influences of this study to liner shipping literature are fourfold. First, it con-

tributes to the literature by developing a realistic liner shipping network design problem

under seasonal demand fluctuations. Second, a mixed integer programming model is

developed for the proposed liner shipping network design problem. Third, a forecasting

model is constructed for estimation of ports under limited historical information. Fourth,

a simulation optimization framework is developed to help decision makers in evaluating

their strategic and tactical level service network decisions.

5. The Feeder Service Network Design 68

5. The Feeder Service Network Design Problems

The Feeder Service Network Design (FND) problem is mathematically modeled in four

parts in this chapter. While Section 5.1 handles the problem in aspects of vehicle rout-

ing problem, Section 5.2 handles the problem as feeder containership routing problem.

Section 5.3 handles the basic FND problem for a stable sailing horizon for reducing the

total transportation cost and Section 5.4 approaches the problem more realistically by

considering varying forecasted throughput demands for a dynamic sailing season and

vessel charter operations.

5.1 The vehicle routing problem with simultaneous pickup and deliv-

ery with time limit

The problem considered in this study is designing the network of service vehicles, i.e.

simultaneously dispatching/collecting cargo parcels from a central post station to/from

regional post stations via trucks, simultaneously dispatching/collecting containers from

a hub port to/from feeder ports via containerships, simultaneously dispatch-

ing/collecting passengers from a continental center airport to/from national airports via

airplanes, etc.

In this context, the vehicle routing problem with simultaneous pickup and delivery

with time limit (VRPSPDTL) a variant of Vehicle Routing Problem (VRP) can be stated

as follows: A set of clients is located on a distribution network where clients require

both delivery and pickup operations. Each client has to be served once for both opera-

tions with a given fleet of identical capacitated vehicles. Each vehicle leaves the central

depot carrying the total amount of goods that it has to deliver and returns to the central

depot carrying the total amount of goods that it must pick-up. Each client also has a

specified service time which is the loading and unloading operation time of the vehicle

at the client. Therefore, the voyage time of a vehicle is the sum of total travel time of the

route and total service time of the clients. In order to determine the vehicle schedules

and the staffing balance, each vehicle has to finish its voyage before the maximal al-

lowed duration is reached.

5. The Feeder Service Network Design 69

This problem can be classified by using the notation offered by Berbeglia et al.

(2007). They used a tuple notation of

[ | | ]Structure Visits Vehicles

for PDP. In this nota-

tion, structure represents the number of origins and destinations of goods, visits repre-

sents information on the way pickup and delivery operations are performed at clients,

vehicles represents the number of vehicles used in the problem. In this content, the

VRPSPD is stated as

[1 1 | | ]M PD m

; where

11M

shows one-to-many-to-one

problems, goods are initially available at the depot and are transported to clients, and

goods available at the clients are transported to the depot;

represents each client

being exactly once for a combined pickup and a delivery operation;

represents that

the solutions could contain more than one vehicle (multi vehicle).

A mixed-integer linear programming (MILP) formulation for the VRPSPDTL has

been presented with the following notation by extending VRPSPD formulation of

(Montane and Galvao 2006):

Indices:

, i j N

the set of clients (0 represents the depot)

kK

the set of vehicles (

KN

)

Parameters:

Maximum allowed voyage duration of vehicles

maximum loading capacity of a vehicle

average travel speed of a vehicle

distance between client

and

service time at client

delivery goods demand of client

pick-up goods demand of client

Decision variables:

1, if the arc

 

,ij

belongs to the route served by vehicle

;

0, otherwise.

pick-up goods transported on arc

 

,ij

delivery goods transported on arc

 

,ij

5. The Feeder Service Network Design 70

The model formulation is given as follows:

ij ij

k K i N j N

min c x

  



(5.1)

s.t.

 

1, j N/ 0

i N k K



  



(5.2)

0, j N,k

ij ji

i N i N

x x K



    



(5.3)

 

x k K



  



(5.4)

 

x k K



  



(5.5)

 

, / 0

ji ij j

i N i N

y y p j N



   



(5.6)

 

, / 0

ij ji j

i N i N

z z d j N



   



(5.7)

, , j N,k

ij ij ij

y z Qx i K    

(5.8)

 

ij kk

ij i ij

i N j N i N j N

cx s x R k K

   

   

  

(5.9)

1 , /0, 2

x k K B NB B



  



(5.10)

 

0,1 , 0, 0, , ,

ij ij ij

x y z i j N k K     

(5.11)

The objective function (5.1) aims to minimize the total travelled distance. Equation

(5.2) ensures that each client is served by only one vehicle; equation (5.3) guarantees

that the same vehicle arrives at and departs from each client. Restrictions (5.4) and (5.5)

ensure usage of maximum K vehicles. Equations (5.6) and (5.7) satisfy pick-up and de-

livery demands of the clients, respectively. Restrictions (5.8) are the vehicle capacity

constraint; restrictions (5.9) represent the maximum voyage duration constraint. (5.10)

are the vehicle sub-tour elimination constraints according to Karlaftis et al. (2009). Fi-

nally, constraints (5.11) define the variable domains. In general, the constraints ensure

that each vehicle departs from the central depot with a load equivalent to the total deliv-

5. The Feeder Service Network Design 71

ery goods and each vehicle arrives at the central depot with a load equivalent to the total

pick-up goods from clients in the route served by that vehicle (See Section 7.1 for the

implementation of the model).

5.2 The feeder containership routing problem

A similar problem is handled by Karlafits et al. (2009) as a feeder containership routing

problem (FCRP) between Aegean Island feeder ports and Greek mainland hub port. In

this context, a set of feeder ports is located on a distribution network where feeder ports

require both delivery and pickup container operations. Each feeder port has to be served

once for both operations with a given fleet of identical capacitated containerships. Each

ship leaves the regional hub port carrying the total amount of container it has to deliver

and returns to the hub port carrying the total amount of containers it must pick-up.

However, while VRPSPDTL has to finish its voyage before the maximal allowed dura-

tion, FCRP aims to deliver its delivery containers before predefined time deadlines

(Karlaftis et al. 2009). Since maritime transportation and feeder port operations are

highly influenced by weather conditions, FCRP uses this time deadline as soft time limit

constraint. Contrary to VRPSPDTL, the routes could violate this time deadlines, how-

ever such routes are penalized some percent of the delays. Moreover, unlike the classi-

cal vehicle routing problems, FCRP aims to minimize overall container voyage time.

According to the definitions above, a mixed integer linear programming (MILP)

formulation for the FCRP could be represented with the following notation:

Indices:

, i j N

the set of ports (0 represents the hub port)

kK

the set of containerships (

KN

)

Parameters:

Service deadline time for feeder ports

maximum loading capacity of a containership

average travel speed of a containership

distance between port

and

service time at feeder port

5. The Feeder Service Network Design 72

delivery container demand of feeder port

pick-up goods demand of feeder port



percentage indicating total time penalty for delays

Decision variables:

1, if the arc

 

,ij

belongs to the route served by containership

;

0, otherwise.

pick-up containers transported on arc

 

,ij

delivery containers transported on arc

 

,ij

Delay in reaching feeder port

for containership

The model formulation is given as follows:

   

ij kk

ij i j

k K i N j N i N k K j N

min x s d



     



   

(5.12)

s.t.

 

1, j N/ 0

i N k K



  



(5.13)

0, j N,k

ij ji

i N i N

x x K



    



(5.14)

 

x k K



  



(5.15)

 

x k K



  



(5.16)

 

, / 0

ji ij j

i N i N

y y p j N



   



(5.17)

 

, / 0

ij ji j

i N i N

z z d j N



   



(5.18)

, , j N,k

ij ij ij

y z Qx i K    

(5.19)

   

/ 0 / 0

ij kk

ij i ij

i N j N i N j N

cx s x R k K

   

   

   

(5.20)

5. The Feeder Service Network Design 73

 

, j N,k

ij kk

ij j

i N j N

L x d K



    



(5.21)

1 , /0, 2

x k K B NB B



  



(5.22)

 

0,1 , 0, 0, 0 , , ,

ij ij ij

x y z i j k Kd N      

(5.23)

The objective function (5.12) aims to minimize the sum value of total voyage dura-

tion (route duration + service time) and penalty of total delay. Equation (5.13) ensures

that each feeder port is served by only one ship; equation (5.14) guarantees that the

same ship arrives at and departs from each feeder port. Restrictions (5.15) and (5.16)

ensure usage of maximum K ships. Equations (5.17) and (5.18) satisfy pick-up and de-

livery container demands of the feeder ports, respectively. Restrictions (5.19) are the

vehicle capacity constraint; restrictions (5.20) represent service time deadline constraint.

Equation (5.21) gives time delays of the ships to feeder ports. (5.22) are the vehicle sub-

tour elimination constraints. Finally, constraints (5.23) define the variable domains. In

general, the constraints ensure that each ship departs from the hub port with a load

equivalent to the total delivery containers and each ship arrives at the hub port with a

load equivalent to the total pick-up containers from feeder ports in the route served by

that ship (See Section 7.2 for implementation of the model).

5.3 Feeder service network design problem

However, so far none of the feeder service network design studies has considered de-

tailed variable and fixed cost structures for a heterogeneous vessel fleet over a complete

sailing horizon. In this study, we propose a mixed-integer linear programming model to

simultaneously determine the fleet size and mix, fleet deployment, ship routing and ship

scheduling in feeder service network design by minimizing total network costs in a sail-

ing season.

In a feeder network, ships visit a number of ports along the predefined routes con-

necting ports in the region. In the design of the feeder network, service factors such as

the capacity of feeder ships, characteristics of the ports, container demand volume at the

various ports as well as bunker costs and operating and chartering costs of the ships

have to be considered. Specifically, the feeder network design problem (FND) can be

described as follows. A set of feeder ports is located on a distribution network where

5. The Feeder Service Network Design 74

feeder ports require both delivery and pickup operations. Each feeder port has to be

served once for both operations with a given fleet of capacitated heterogeneous feeder

ships. Each ship leaves the hub port carrying the total amount of containers it has to

deliver and returns to the hub port carrying the total amount of containers picked up on

the voyage. Each feeder or hub port also has a specific operation efficiency for loading

and unloading containers. The service time of the ports depends on the port operation

efficiency, ship size, the number of loaded and unloaded containers and the pilotage

time for entering and exiting the port. Therefore, the total voyage duration of a ship

consists of the total travel time of the route and the total service time at the hub and the

feeder ports. The voyage starts in the hub port with commencing the loading operations

to ships and completing the unloading operations from ships at the hub port. Each vessel

has to finish its voyage before the allowed time deadline is reached. Before starting a

new voyage, the ship needs a lay-up interval for repair, cleaning, waste disposal etc.

According to these considerations the FND problem has similarities with the “vehi-

cle routing problem with simultaneous pick-up and delivery with time limit”

(VRPSPDTL). For a review and classifications of vehicle routing problems, see e.g.

Berbeglia et al (2007) and Parragh et al (2008). While the VRPSPDTL aims to mini-

mize total voyage distance and FCRP aims to minimize total voyage duration, the FND

problem aims to serve all contracted feeder ports by minimizing total operational costs

for a sailing season. For a feeder network provider, operational costs of the planning

period include containership related fixed costs for the necessary number of ships (char-

tering, operating etc.) and total service related variable costs (on-sea bunker cost, on-

port bunker cost, port set-up charges). Table 5.1 shows the related basic cost calcula-

tions.

Table 5.1: Basic calculations of total costs for a sailing season

Parameter

Basic calculation

Total costs

Fixed costs + Variable costs

Fixed costs

Number of necessary ships (Chartering + Operating costs)

Variable costs

Number of services * (Bunker (sea) + Bunker (port) + Port set up charges)

Number of required ships

(Voyage duration + Lay up duration) / Service frequency

Number of services

Planning period / Service frequency

Voyage duration

On-sea duration + On-port duration (feeder) + On-port duration (hub)

Idle duration

Number of necessary ships Service frequency - (Voyage + Lay-up duration)

Ship total duration

Voyage duration + Lay up duration + Idle duration

5. The Feeder Service Network Design 75

Since our investigation is concerned with the design of a real world container feeder

network, some assumptions have to be made in order to exclude elements of minor rele-

vance and to focus on those aspects that are of paramount interest.

 All parameter values are deterministic, i.e. we exclude weather and seasonal ef-

fects, for instance.

 No direct delivery takes place between feeder ports.

 The queue time at ports is not considered.

 Time windows for berthing at a port are disregarded since they are not known in

advance for a complete sailing season.

 Demand and supply quantities of feeder ports cannot be split.

 The ships of a certain type are identical regarding their carrying capacity.

 Bunker costs are the same in all ports.

 We only consider chartered ships.

 Set-up durations (pilotage, berthing, cleaning etc.) of a port only depend on the

type of ship.

 Effects of speed dependent fuel costs as well as straight/canal durations and

costs are not considered.

A mixed-integer linear programming (MILP) formulation of the FND problem is

presented using the following notation:

Indices & sets

, i j N

The set of ports (0 represents the hub port)

sS

The set of containership types

 

, i j L

The set of allowed voyage legs between ports

rR

The set of routes (

RN

)

Parameters

Service frequency

days



Number of services in a sailing season

Sailing season duration

days

5. The Feeder Service Network Design 76

Maximum allowed voyage duration

hours

Vessel set-up duration of ship type

in port

(pilotage,

berthing, cleaning etc.)

hours

Lay-up duration of ship type

Hours

Available number of containerships of ship type

ships

Loading capacity of ship type

TEU

Average travel speed of ship type

n.mile/hour

Operation efficiency of port

for ship type

TEU/hour

Distance between ports

and

n.mile

Berthing duration of ship

at port

Hours

Daily container demand (delivery) of port

TEU

Daily container supply (pick-up) of port

TEU



Main fuel oil price

$/ton



Auxiliary fuel oil price (distillate)

$/ton

Chartering cost of ship type

$/ship

Operating cost of ship

(administration, maintenance, lubri-

cant, insurance etc.)

$/ship

Main fuel consumption of ship type

on sea

ton/n.mile

Auxiliary fuel consumption of ship type

at berth

ton/hours

Vessel set-up cost of ship type

at port

$/ship

Decision variables

1, if the arc between ports

and

belongs to route

served

by ship type

(0, otherwise)

Binary

containers picked up from ports up to port

and transported

from port

Integer

containers to be delivered to ports routed after port

and

transported between port

and

Integer

Required number of ships of type

on route

Ships

Voyage cycle time of route

with ship type

Hours

Total fixed costs of a sailing season

5. The Feeder Service Network Design 77

Total variable costs of a sailing season

The MILP model formulation is given as follows:

minFC VC

(5.24)

s.t.

 

s s rs

r R s S

FC cc fc e



  



(5.25)

 

 

s s / 0 s

s rs s s s rs

ij ij i i ij

i N j N r R S i N S i N j N r R S

VC w a x t b bc x

p d f

with and t

  



         



   













     

    

(5.26)

rs s rs

cue r R S

   

(5.27)

rs s s rs

i i ij

i N j N

c t v x r R S





      







(5.28)

 

1 N/ 0

i N r R S

  

  



(5.29)

0 N, ,s

rs rs

ij ji

i N i N

x x j r R S



     



(5.30)

 

1 ,

rsj

x r R s S



   



(5.31)

 

1 ,

x r R s S



   



(5.32)

1 , , /0, 2

i B j B

Bx r r s S B N B



    



(5.33)

rs s

i N j N r R

x m s S

  

  



(5.34)

ji ij j

i N i N

y y p f j N



   



(5.35)

ij ji j

i N i N

z z d f N



   



(5.36)

, , ,

s rs

ij ij ij

y z q x i j N r R s S     

(5.37)

5. The Feeder Service Network Design 78

c K r R s S   

(5.38)

   

0,1 , , , , , , , 0 ,,

rs rs rs

ij ij ij

x y z c i j N i je L r R s S



      

(5.39)

The objective function (5.24) minimizes total costs of the network for a sailing sea-

son. Equations (5.25) and (5.26) define fixed and variable costs, respectively. The nec-

essary number of ships needed for a full service cycle on each route is calculated in

(5.27). Equation (5.28) determines the cycle time of ships on each route (berthing dura-

tion + service duration + voyage duration). Equation (5.29) ensures that each feeder port

is served by only one type of ship and one route. Equation (5.30) guarantees that a ship

arrives at and departs from each feeder port on each route. (5.31) and (5.32) impose a

similar condition for the hub port at which the route starts and ends. (5.33) are the vehi-

cle sub-tour elimination constraints according to Karlaftis et al (2009). Constraints

(5.34) represent an upper bound for the number of ships employed from each type.

Equations (5.35) and (5.36) satisfy pick-up and delivery demand of containers at the

feeder ports, respectively. (5.37) are the ship capacity constraints. (5.38) represent the

maximum voyage duration constraints. Finally, constraints (5.39) define the variable

domains. In general, the constraints ensure that each ship departs from the hub with a

load equivalent to the total delivery of containers and each ship returns to the hub with a

load equivalent to the total pick-up containers from feeder ports in the route served by

that ship (See Section 7.3 for implementation of the model).

5.4 Liner shipping network design under unstable demand environ-

ments

In most existing studies, liner shipping service networks are presented under the as-

sumption that the container supply and demand of ports is given only as a set of the sta-

ble values of sailing season. The studies considering demand seasonality and fluctua-

tions are interested with the problems in already designed networks (see Section 4.5).

However, so far none of the liner shipping service studies have considered the effect of

seasonal demand fluctuations on the design of service networks over a complete dynam-

ic sailing horizon. In this study, we propose a mixed-integer linear programming model

to simultaneously determine the fleet size and mix, fleet deployment, ship routing and

ship scheduling in H&S feeder service network design by minimizing total network

costs under seasonal demand fluctuations over periods of a sailing season. The proposed

5. The Feeder Service Network Design 79

model periodically allows changing service network designs, routes, fleet deployments

and schedules according to forecasted demand of sailing season periods determined by

the decision makers of the liner shipping industries.

The MILP formulation proposed in Section 5.3 is extended by considering unstable

demand environments of ports over periods for a sailing season, including the ships on

hand, and allowing unnecessary ships to charter out. An extended MILP formulation of

the LSND problem is presented using the following notation:

Indices & sets

, i j N

The set of ports (0 represents the hub port)

sS

The set of containership types

 

, i j L

The set of allowed voyage legs between ports

rR

The set of routes (

RN

)

gG

The set of allowed network change periods

Parameters

Service frequency

days

Duration of period

days



Number of services in period

Maximum allowed voyage duration

hours

Vessel set-up duration of ship type

in port

(pilotage,

berthing, cleaning etc.)

hours

Lay-up duration of ship type

Hours

Available number of containerships of ship type

for char-

ter in

ships

On hand number of ship type

ships

Loading capacity of ship type

TEU

Average travel speed of ship type

n.mile/hour

Operation efficiency of port

for ship type

TEU/hour

Distance between ports

and

n.mile

Berthing duration of ship at port

in period

Hours

5. The Feeder Service Network Design 80

Container demand (delivery) of port

in period

TEU/days

Container supply (pick-up) of port

in period

TEU/days



Main fuel oil price

$/ton



Auxiliary fuel oil price (distillate)

$/ton

Charter in cost of ship type

$/day

Charter out price of ship type

$/day

Owning cost of ship type

$/day

Operating cost of ship (administration, maintenance, lubri-

cant, insurance etc.)

$/day

Main fuel consumption of ship type

on sea

ton/n.mile

Auxiliary fuel consumption of ship type

at berth

ton/hours

Vessel set-up cost of ship type at port

$/ship

Decision variables

gij

1, if the arc between ports

and

belongs to route

served

by ship type

in period

(0, otherwise)

Binary

gij

containers picked up from ports up to port

and transported

from port

in period

TEU

gij

containers to be delivered to ports routed after port

and

transported between port and

in period

TEU

Number of necessary ships from type

on route

in period

Ships

Voyage cycle time of route

with ship type

in period

Hours

Number of used owned ships on route

from type

in peri-

Ships

Number of charter out ships from type

in period

Ships

Number of used charter in ships on route

from type

period

Ships

Total fixed costs of a sailing season

Total variable costs of a sailing season

The MILP model formulation is given as follows:

5. The Feeder Service Network Design 81

minFC VC

(5.40)

s.t.

   

s s rs s s rs s s

g g g

r R s S g B r R s S g G s S g G

FC oc fc su cc fc ci co cp

       

    

  

(5.41)

 

 

s s / 0 s

rs s s s rs s

g ij gij g gi g gij i

g G g G g G

i N j N r R S i N S i N j N r R S

gi gi

g gi s

VC w x t af x bc

p d f

with

and t

    



  

         

  









  

    

(5.42)

, ,

rs s

grs

cue r R s S g G

    

(5.43)

, ,

s s rs rs

gi i gij g

i N j N

t v x c r R s S g G





     









(5.44)

 

1 / 0 ,

gij

i N r R s S

x j N g G

  

   



(5.45)

0 , , ,

rs rs

gij gji

i N i N

x x j N r R s S g G



      



(5.46)

 

1 , ,

x r R s S g G



    



(5.47)

 

1 , ,

x r R s S g G



    



(5.48)

, ,

rs rs rs

g g g

su ci r R s S g Ge 

(5.49)

rs s s

su co sn s S g G



 



(5.50)

m s S gi Gc



   



(5.51)

gji gij gj

i N i N

y y p f j N g G



    



(5.52)

gij gji gj

i N i N

z z d f j N g G



    



(5.53)

, , , ,

s rs

gij gij gij

y z q x i j N r R s S g G      

(5.54)

, ,

c K r R s S g G    

(5.55)

5. The Feeder Service Network Design 82

1 , , , /0, 2

gij

i B j B

x r r s G B BB S g N



     



(5.56)

 

 

0,1

, , , ,, , ,,

,,,

rs s rs

gij

gij gij g

su co ci

y z e i j N i j L r R s S g G





      



(5.57)

The objective function (5.40) minimizes total costs of the network for a sailing sea-

son. Equations (5.41) and (5.42) calculate fixed and variable costs, respectively. The

necessary number of ships needed for a full service cycle on each route is calculated in

(5.43). Equation (5.44) calculates the cycle time of ships on each route (berthing dura-

tion + service duration + voyage duration). Equation (5.45) ensures that each feeder port

is served by only one ship and equation (5.46) guarantees that the same ship arrives at

and departs from each feeder port on each route. (5.47) and (5.48) impose a similar con-

dition for the hub port at which the route starts and ends. Equations (5.49) and (5.50)

calculate the number of charter in and charter out ships, respectively. Constraints (5.51)

represent the maximum number of available charter in ships employed from each type.

Equations (5.52) and (5.53) satisfy pick-up and delivery demand of containers at the

feeder ports, respectively. Equations (5.54) are the ship capacity constraints. Re-

strictions (5.55) represent the maximum voyage duration constraint. (5.56) are the vehi-

cle sub-tour elimination constraints. Finally, constraints (5.57) define the variable do-

mains. In general, the constraints ensure that each ship departs from the hub with a load

equivalent to the total delivery of containers and each ship returns to the hub with a load

equivalent to the total picked-up containers from feeder ports in the route served by that

ship.

In this study, the service network is revised at the beginning of every period in re-

sponse to changes in demand patterns for a season. Changes to the service network may

include introducing new routes, and schedules as well as fleet deployments which could

contain chartering in new ships or chartering out unnecessary ships. In contrast to trunk

liners, feeder service providers have the ability to perform frequent and eﬃcient updates

to the schedule and routes with the help of fleet deployments. This study also employs

various service scenarios in order to better help decision makers of liner shipping pro-

viders. These scenarios contain different periodical season approaches, different de-

mand allocations, different numbers of owned ships at the start of sailing season, differ-

5. The Feeder Service Network Design 83

ent ship owning, chartering, and oil prices to provide a very high degree of ﬂexibility to

planning decisions (See Section 7.4 for implementation of the model).

6. The Proposed Solution Methodology 84

6. The Proposed Solution Methodology

This chapter provides two frameworks for the feeder service network design problems.

The first framework (Section 6.1) proposes a novel approach to optimally solve related

feeder service network problems and second framework (Section 6.2) provides a simu-

lation based forecasting approach in order to estimate seasonal demand fluctuations in

feeder service.

6.1 The adaptive neighborhood search approach

The network design problems presented in the previous section is a highly complex

combinatorial optimization problem and thus hard to solve by use of standard optimiza-

tion software. Exact methods for solving the network design problems are generally not

practical for large instances because of the problem complexity. Therefore the model is

not intended for solving the mathematical models. In this study, we propose a novel

adaptive neighborhood search (ANS) algorithm based on heuristic approaches. The

steps of the approach are described in Figure 6.1.

The algorithm applies the Savings Algorithm (SA) in order to gain a fast and effec-

tive initial solution. The ANS is embedded with Variable Neighborhood Search (VNS)

to improve the initial solution by searching neighborhoods. In order to escape from lo-

cal optima, an Adaptive Perturbation Mechanism (APM) is developed.

6.1.1 Saving heuristics

The proposed ANS approach applies the Savings Algorithm of Clarke and Wright

(1964) in order to gain a fast and effective initial solution. This classic heuristic aims at

merging sub-tours based on cost savings which can be achieved by combining two sub-

tours to be served by one vehicle (see Figure 6.2). In the literature, some enhancements

of the Clarke and Wright savings algorithm have been suggested by adding new terms

and parameterizing the savings formula. Since the VRPSPDTL problem is a generaliza-

tion of the vehicle routing problem (VRP), we construct our initial solution by extend-

ing the savings formula proposed for the VRP by Altinel and Öncan (2005).

6. The Proposed Solution Methodology 85

procedure:ANS approach

input: parameters and structures, kmax, mmax, nmax, smax, pmax

output: 𝜋

start

construct 𝜋0; {construct an initial solution with savings heuristic}

𝜋1← 𝜋0,𝑝 ← 1, ← 1;

repeat

𝑘 ← 1

10:

repeat

11:

𝜋2← 𝜋𝑘

1; {shaking- 𝜋𝑘

1 is a random solution in the kth neighborhood of 𝜋1,𝑘 ∈ 𝑁𝑘 }

12:

if 𝑓(𝜋2) < 𝑓(𝜋1)

13:

𝜋1← 𝜋2,𝑘 ← 1;

14:

else

15:

𝜋3← 𝜋2,𝑚 ← 1, 𝑛 ← 1;

16:

repeat

17:

repeat

18:

𝜋4← 𝜋𝑚

2; {local search- 𝜋𝑚

2 is a random sol. in the mth n.hood of 𝜋2,𝑘 ∈ 𝑁𝑚 }

19:

if 𝑓(𝜋4) < 𝑓(𝜋3)

20:

𝜋3← 𝜋4;

21:

end

22:

𝑚 ← 𝑚 + 1;

23:

until m= mmax

24:

𝑛 ← 𝑛 + 1, 𝑚 ← 1;

25:

until n= nmax

26:

if 𝑓(𝜋3) < 𝑓(𝜋1)

27:

𝜋1← 𝜋3,𝑘 ← 1;

28:

else

29:

𝑘 ← 𝑘 + 1;

30:

end

31:

end

32:

until k= kmax

33:

if 𝑓(𝜋1) < 𝑓(𝜋0)

34:

𝜋0← 𝜋1; {move or not -𝜋0 is current best solution }

𝑝 ← 1, ← 1;

else

← + 1;

end

until s= smax

𝜋1← 𝜋𝑟

0,;{ APM- 𝜋𝑟

0 is a random solution in the random rth n.hood of 𝜋0,𝑟 ∈ 𝑁𝑟 }

𝑝 ← 𝑝 + 1, ← 1;

until p= pmax

𝜋 ← 𝜋0

44:

end

* k-max: number of shaking structures of VNS; m-max: number of local search structures of VNS; n-max: number of local search

repetition; s-max: termination number of ANS; p-max: perturbation call number of APM;: Nk: the set of shaking neighborhood

structures; Nm: the set of local search neighborhood structures; Nr: the set of perturbation neighborhood structures.

Figure 6.1: Structure of the ANS approach

The savings formula is given in Equation (6.1) where

is the distance of customer

i to the depot,

is the distance of the depot to customer j, and

is the distance be-

tween customers i and j,

and

are the demand of customer i and j,

is the average

demand.

6. The Proposed Solution Methodology 86

0 0 0 0

ij i j ij i j

S c c c c c d

  



     

(6.1)

Here the First positive parameter λ aims to redesign the routes in order to ﬁnd better

solutions. Second positive parameter μ may exploit the asymmetry information between

customers i and j regarding their distances to the depot. Third positive parameter ν gives

an assignment priority to customers with larger demands (Doyuran and Çatay 2011).

Since this savings function is designed for the VRP, we assume

as maximum

value of demand

 

and pick-up

 

of customer i

 

' max ,

i i i

d d p

in the

VRPSPDTL. This assumption converts the VRPSPDTL into the vehicle routing prob-

lem with time limit (VRPTL). After an initial solution is constructed with the savings

heuristics for the VRPTL, this solution is evaluated with an improvement algorithm

according to the VRPSPDTL.

Step 1: Calculate the savings s(i, j) for every pair (i, j) of customers.

Step 2: Rank the savings s(i, j) in descending order. This creates the "savings list." Process the

savings list beginning with the topmost entry in the list (the largest s(i, j)).

Step 3: For the savings s(i, j) under consideration, include link (i, j) in the route, if no route

constraints (vehicle capacity, route and time limit) will be violated through the inclusion of

(i, j), and if:

a) neither i nor j have already been assigned to a route, in which case a new route is

initiated including both i and j,

b) or, exactly one of the two customers (i or j) has already been included in an

existing route and that customer is not interior to that route in which case the link (i,

j) is added to that same route,

c) or, both i and j have already been included in two different existing routes and

neither customer is interior to its route, in which case the two routes are merged.

Step 4: If the savings list s(i, j) has not been exhausted, return to Step 3, processing the next entry

in the list; otherwise, stop: the solution to the VRPTL consists of the routes created.

Step 5: Any customer that has not been assigned to a route during Step 3 must be served by a

vehicle that begins at the depot visiting the unassigned customer and returning to depot.

Figure 6.2: Structure of construction heuristic (Kulak et al. 2011)

6.1.2 Variable neighborhood search

In the next stage, the initial solution is improved with Enhanced Variable Neighborhood

Search (EVNS). The EVNS is an adapted version of the Variable Neighborhood Search

(VNS) approach of Mladenović and Hansen (1997). VNS is based on the idea of sys-

tematically changing the neighborhoods in order to improve the current solution and

aims to explore the solution space which cannot be searched by local search (Hansen et

al. 2010). Kytöjokia et al. (2007), Hemmelmayr et al. (2009) and Stenger et al. (2012)

6. The Proposed Solution Methodology 87

showed the effectiveness of VNS in VRP applications. Shaking, local search and move

or not operators are used in the implementation of the VNS. The shaking operator de-

fines the search direction of the VNS by using the set of neighborhoods. The possibility

of reaching a global solution increases when combining the shaking operator with local

search rather than using a single shaking operator. Therefore, each solution obtained

through the shaking operator is used in the local search operator in order to explore new

promising neighborhoods of the current solution. In this study we implemented the Var-

iable Neighborhood Descend (VND) algorithm as the local search operator. The VND

aims to combine the set of neighborhoods (m-max) in a deterministic way, since using

more than one neighborhood structure could obtain a better solution. After each shaking

operation, the VND algorithm allows n-max trials for maximum possible improvement.

At the end of the VND algorithm, if there is an improvement, then the shaking opera-

tions start from the first operation, if not, the shaking continues with the next operation.

After reaching the maximum number of shaking operations (k-max), the search contin-

ues with the first operation in the new iteration (Hansen and Mladenović 2001).

In this study, a set of neighborhood structures [3-opt, swap, insertion, 2-opt, Ex-

change (m,n), Cross, Shift (0,1), Replace (1,1)] is employed in a deterministic order as

shaking and local search operators (Figure 6.3). To avoid redundant moves, only moves

under violation acceptance limits are admitted in the shaking operator. The total route

duration violation acceptance limit

 



is used to allow clients to join another route for

possible future improvements. Also the vehicle capacity violation acceptance limit

 



is used as the maximum of all pick-up and delivery loads. However, just one of the

routes is allowed to use this violation acceptance limit and the travel duration of this

route is punished with a huge penalty cost in order to increase the improvement possi-

bility of routes in the local search phase. In the local search phase, only feasible move-

ments are admitted, i.e. those which do not violate the ship capacity and time limit. Al-

so, reverse routes are checked in terms of a capacity violation.

The 3-opt, swap, insertion and the 2-opt are intra-route neighborhood structures de-

fined according to an initial configuration (see Figure 6.3.a below).

The 3-opt, which was introduced by Lin (1965), tries all shifts of some sub-

sequence to different positions in the same route. Specifically, three edges are deleted

and replaced by three other edges. The links [1,2] (between customer 1 and 2), [3,4],

6. The Proposed Solution Methodology 88

[5,6] were deleted from route 1 (Figure 6.3.a) and the links [1,4], [5,2] and [3,6] were

inserted (Figure 6.3.b).

3-opt

Cross

Initial Configuration

Swap

Shift (0,1)

b g

d i

c h

Insertion

Replace (1,1)

e j

Exchange (m,n)

Initial Configuration

2-opt

a f

Figure 6.3: Neighborhood structures

The swap is a random permutation movement between two customers in the same

route. The route order of customer 2 and customer 5 were swapped (Figure 6.3.c.)

Insertion operation selects a customer randomly and inserts it in a random position

in the same route. Customer 3 was selected and inserted at the 6th position of the route

(Figure 6.3.d)

6. The Proposed Solution Methodology 89

The 2-opt heuristic looks for improvements by swapping pairs of links (Croes

1958). Links [1,2] and [4,5] were deleted and links [1,4] and [2,5] were inserted (Figure

6.3.e).

Exchange (m,n), Cross, Shift (0,1) and Replace (1,1) are inter-route structures de-

fined according to an initial configuration (see Figure 6.3.f).

Exchange (m,n) structure shown in Figure 6.3.g. is developed according to the idea

of Osman (1993). In this figure, m sequential customers from one route (route 1) are

transferred to another route (route 2) and n sequential customers from route 2 are trans-

ferred to route 1. In this study, m is randomly selected between 1 and 5, and n is ran-

domly selected equal to m or one lower.

Cross exchange is a basic crossover structure between routes. In this structure, the

link [2,3] from route 1 and the link [7,8] from route 2 are removed. Later, links [2,8]

and [7,3] were inserted (Figure 6.3.h).

Shift (0,1) is a random transposition movement of a customer from one route to an-

other. Customer 2 from route 1 was transferred to route 2 (Figure 6.3.i).

Replace (1,1) is a random permutation movement between two customers from dif-

ferent routes. Customer 1 from route 1 is permutated with customer 7 from route 2

(Figure 6.3.j).

In all inter-route neighborhood structures, the route pairs are selected according to

the roulette wheel in order to eliminate the number of infeasible exchange operations. In

this selection, the center of gravity coordinates of each route is calculated. After one

route is randomly selected, the distances between center of this route and the center of

other routes are calculated. All distances between the selected route and remaining

routes are scaled by a

0.5

1distance

factor in order to give exchange opportunity to dis-

tant routes. Another route is selected from these routes according to principles of rou-

lette wheel method.

6.1.3 Adaptive perturbation mechanism

The temporary solution which is obtained via the shaking and local search operators is

compared with the current solution in order to decide whether to move or not. In the

6. The Proposed Solution Methodology 90

proposed VNS and VND, the acceptance criterion of the temporary solution accepts

only improvements. However, this procedure may cause the search to stuck in a local

optimum. Therefore, it is necessary to employ a strategy of accepting non-improving

solutions. Perturbation is an effective strategy used to jump out of a local optimum and

to search a new promising region. A commonly used perturbation strategy is to destruct

the previous local optimum partially in a random way (Subramanian et al. 2010). An-

other strategy is the destroy-and-repair based perturbation mechanism (Jun and Kim

2012).

The previously obtained local optimum solution combines global statistical infor-

mation and local information of good individual solutions. In this study, the current so-

lution is there-fore used to develop a novel perturbation method called Adaptive Pertur-

bation Mechanism (APM). This perturbation mechanism runs after a number of non-

improving iterations counted from the last improving iteration (p-max). In addition to

the perturbation move, a local optimization method with the previously defined four

intra-route neighborhood structures is applied in order to improve the perturbed solution

quality (see Figure 6.4 for the steps of the algorithm).

procedure:ANS approach

input: parameters and structures, ℎ𝑚𝑎𝑥 , 𝑧𝑚𝑎𝑥 , 𝜋0, AL

output: 𝜋1

start

𝜋1← 𝜋0+𝐴𝐿;

repeat

𝑧 ← 1;

𝜋5← 𝜋𝑟

0; {perturb- 𝜋𝑟

0 is perturbed with a random perturbation structure 𝑟 𝑟 ∈ 𝑁𝑟 }

repeat

10:

ℎ ← 1;

11:

repeat

12:

𝜋6← 𝜋ℎ

5;{ optimization- 𝜋ℎ

5 is a random sol. in the h th n.hood of 𝜋5,ℎ ∈ 𝑁ℎ};

13:

if 𝑓 𝜋6 <𝑓 𝜋5

14:

𝜋5← 𝜋6,𝑧 ← 1, ℎ ← 1, 𝑦 ← 1;

15:

else

16:

ℎ ← ℎ + 1;

17:

end

18:

until ℎ=ℎ𝑚𝑎𝑥

19:

𝑧 ← 𝑧 + 1;

20:

until 𝑧=𝑧𝑚𝑎𝑥

21:

if 𝑓 𝜋5 <𝐴𝐿

22:

𝜋1← 𝜋5;

23:

end

24:

until 𝜋1<𝐴𝐿

25:

end

* h-max: number of optimization structures; z-max: max number of perturbation method attempts; Nr: the set of perturbation struc-

tures; Nh: the set of optimization structures; AL: acceptance limit of the route.

Figure 6.4: Structure of APM algorithm

6. The Proposed Solution Methodology 91

In the APM, a set of perturbation structures [double replace, double cross, triple

shift, triple replace, and triple cross] is randomly run whenever the perturbation is

called. In addition to the perturbation move, a local optimization method with previous-

ly defined four intra-route neighborhood structures is applied in order to improve the

perturbed solution quality. The solution quality of the perturbed solution is significant

since a perturbation move that satisfies the vehicle capacity and total route duration lim-

it is always accepted. Moreover, violating moves are accepted as by the shaking opera-

tor. The new developed perturbation structures for the APM are defined as follows.

Double Replace is a combination of two times sequential Replace (1,1) movements

to the same routes which are selected by the roulette wheel method. A random client

from route 1 is permutated with a random client from route 2; next, another random

client from route 1 is permutated with a client from route 2. After intra local search is

applied to both route1 and route 2, the total vehicle duration and vehicle loading capaci-

ty are checked according to the acceptance limits.

Double Cross applies the Cross exchange. Otherwise, it is similar to the Double Re-

place structure.

Triple Shift is a newly developed fast and effective perturbation movement to jump

out from local optima. A route (route 1) is randomly selected, and two another routes

(route 2 and route 3) are selected by using the defined roulette wheel method. Next, sim-

ilar to the Shift (0,1) movement a random client from route 2 is transferred to route 1,

and a client from route 1 is transferred to route 3. Similar to double structures, vehicle

duration and capacity are checked according to the acceptance limit after intra-local

search applied to routes.

Triple Replace (is similar to Triple Shift by using the Replace (1,1) movement.

Triple Cross is similar to Triple Shift by using the Cross exchange structure.

As local optimization, a set of intra neighborhood structures [3-opt, Swap, Insertion,

2-opt] are repeated z-max times in a deterministic order. If no acceptable solution is

generated after z-max attempts in any perturbation structure, the algorithm then tries

another perturbation structure.

6. The Proposed Solution Methodology 92

6.2 Forecasting framework

Forecasting is a method of estimating statements about future events for which actu-

al results have not yet been observed. Forecasting could help decision-makers plan for

the future. Parallel to national trade of countries, container throughputs of regional ports

have high seasonality and demand fluctuations (Schulze and Prinz 2009; Polat and Uslu

2010). Therefore, reliable and accurate forecasting is needed to help the decision makers

plan liner shipping service more effectively and efficiently, since container shipping

involves considerable capital investments and huge daily operating costs. In the litera-

ture, the proposed models for liner shipping feeder service network design problems

consider stable container demand of ports, because of the major complexity of real-

world systems. Therefore, in this study, a simulation and artificial neural networks

based forecasting framework is developed in order to analyze the impact of seasonal

demand fluctuation on the liner shipping feeder service network design.

In developed structure, forecasting a frame consists of three modules (see Figure

6.5). The first decomposition is used to convert yearly maritime statistics to monthly

container throughput information. The second artificial neural network (ANN) module

is used to reflect trend and seasonality in forecasting monthly container throughput and

the third simulation module is used to reflect daily demand fluctuations on container

throughput.

Forecasting Framework

Weekly expected

export and import

throughputs

Yearly

statistical

total

container

throughputs

Decomposition

Monthly

decomposed export

and import

throughputs

Estimation

Monthly forecasted

export and import

throughputs

Simulation

Daily simulated

export and import

throughputs

Figure 6.5: Forecasting framework

6.2.1 Decomposition mechanism

Quantitatively oriented literature and databases on international container throughput is

quite limited (Schulze and Prinz 2009). In addition, shipping lines and container ports

usually just provide yearly market shares and total handling amounts. Therefore, the

decomposition module deconstructs yearly throughputs to monthly supply and demand

amounts. See Section 7.4.1 for implementation of decomposition mechanism.

6. The Proposed Solution Methodology 93

6.2.2 Estimation mechanism

ANNs are computational models inspired by the brain and how it processes information.

Instead of requiring detailed information about the nature of a system, ANNs try to

learn the relationship between the variables and parameters by checking data. ANNs can

also handle very complex and large systems with many interrelated parameters. The

effectiveness of biological neural systems originates from the parallel-distributed pro-

cessing nature of the biological neurons. An ANN simulates this system by distributing

computations to small and simple processing nodes (artificial neurons) in a network.

ANNs have been used in many fields. One major application area is forecasting. Due to

the characteristic features, ANNs are an attractive and appreciated alternative tool for

both forecasting researchers and practitioners. For comprehensive reviews on the appli-

cation of ANNs on forecasting, please see Zhang et al. (1998) and Kline and Zhang

(2004). Therefore, ANNs are a common tool in forecasting container throughputs of

container terminals (See Section 4.4). In developed framework, multi-layer feed- for-

ward networks are trained using back-propagation in order to make estimations for each

port’s monthly demand and supply throughputs.

Figure 6.6 shows typical multi-layer feed-forward ANN architecture. A typical

ANN contains three layers: an input layer, an output layer and, between them, the hid-

den layers. Each artificial neuron (node) is linked to nodes of the previous layer with

weights. A set of these weights creates the knowledge from the system. In order to pro-

duce the desired output for a presented input, the network is trained with a learning

method through adaptation of the weights. After the training operation, the weights con-

tain meaningful information about the data. The network uses the corresponding input

data to produce an output data, which is then compared with the desired output. When

there is a difference between desired and produced outputs, the weights continue to

adapt in order to decrease difference (error). Until the total error reaches the required

limit, the network continues to run in all the input patterns. After reaching the accepta-

ble level, the ANN stops and uses the trained network to make forecasts. For details,

please see Zurada (1992) and Bose and Liang (1996 ) for details of algorithm.

6. The Proposed Solution Methodology 94

...

Hidden layers Output layer

Neuron (node) Weights

Input layer

Figure 6.6: Typical ANN architecture

The back-propagation (BP) is a gradient-descent based effective learning algorithm

for ANNs (Rumelhart et al. 1986). By adapting the weights with the gradient, PB tries

to reduce the total error. The error is calculated with root-mean-square (E) value in

Equation (6.2), where t is produced and o is desired outputs over all patterns (p) and

nodes (i).

1/2

2ip ip

E t o











(6.2)

BP algorithm first assigns random values to all weights in all nodes. Then, the acti-

vation

 



value is calculated for each pattern and for each node by using the activa-

tion function given in Equation (6.3), where j refers to all nodes of the previous layer, i

refers to all node positions of current layer, and

and

are input and weight terms.

pi j ij

f x w











(6.3)

After calculating the output of the layer, the error term

 



for each node is also

calculated back through the network. The error term measures the changes in the net-

work by using changes in the weight values. The error term is calculated for the output

nodes and for the sigmoid activation function as given in Equation (6.4). For hidden

layer nodes, the error term is calculated as given in Equation (6.5), where k indicates

nodes in the downstream layer and j is the position of the weight in each node.

6. The Proposed Solution Methodology 95

   

pi pi pi pi pi

   

  

(6.4)

 

pi pi pi pi kj

   





(6.5)

In conclusion, incremental change to each weight for each node is calculated as giv-

en in Equation (6.6), where ε is learning rate used for weight adaptation in each training

iteration and m is momentum, used to change the weight in the previous training itera-

tion

 

w

. Stopping conditions, maximum iteration number, learning rate and momen-

tum are speed and stability constants defined at the beginning of the training.

   

ij pi pi ij

w m w

  



  

(6.6)

6.2.3 Simulation mechanism

The Monte Carlo simulation module uses the monthly throughputs estimated by the

ANNs module as an input in order to generate daily demand and supply expectations of

container terminals. By analyzing these expectations, shipping line planners can obtain

realistic predictions for slot capacities, network designs, routes and schedules in the

future. Simulation is run many times by using throughput forecasts of ANNs. By the

way, diﬀerent random components of the future demand and supply movements of ports

are obtained for shipping line planners. See Section 7.4.1 for implementation of simula-

tion mechanism.

7. Numerical Investigation 96

7. Numerical Investigation

The Feeder Service Network Design (FND) related problems are solved in four sections

in this chapter. Section 7.1 solves the benchmark instances of the vehicle routing prob-

lem with simultaneous pickup and delivery with time limit (VRPSPDTL), which is

known as the background problem of the FND. This section proof robustness and effec-

tiveness of the developed Adapted Neighborhood Search (ANS) approach. Section 7.2

solves Containership Routing Problem (FCRP) for a case study from the literature. This

problem is the basic version of FND without including sailing season, ship economies,

and ships mix and deployment. This section shows effectively implementation of ANS

to containership routing problems. Section 7.3 solves the FND problem which aims to

simultaneously determine the fleet size and mix, fleet deployment, ship routing and ship

scheduling by minimizing total network costs in a sailing season by using a case study

from Black Sea region with the help of developed ANS approach. Finally, Section 7.4

estimates container weekly container throughput of ports from Black Sea region under

limited historical data with developed simulation and artificial neural networks based

forecasting framework and solves the extended FND problem under seasonal demand

with the help of developed ANS approach. The related section also handles various ser-

vice scenarios such as different periodical approaches, different demand allocations,

different number of owned ships in the starting of season, different ship owning, char-

tering, and oil prices in order to better help decision makers of liner shipping providers.

7.1 VRPSPDTL application

The presented mathematical model in Section 5.1 has been programmed by use of the

GAMS 23.7 software with the CPLEX 12 solver on an Intel Core-2-Duo T5750 2.0

GHz processor with 3 GB RAM. The proposed metaheuristic approach in Section 6.1

has been coded using Matlab R2009a and executed by using Visual C# 2010 on the

same computer.

7. Numerical Investigation 97

7.1.1 Benchmark instances

The performance of presented model and proposed metaheuristic are firstly tested on a

real case study provided by Min (1989). In this study, a library administration center

acts as a depot to 22 client public libraries in a region. The administration center has

two homogenous vehicles with 10500 amount capacities. The total delivery amount

from depot to client libraries is equal to 20300, and the total pickup amount from the

libraries is equal to 19950. Since the original problem is presented as VRPSPD; we

added average vehicle speed parameter as 1 distance/unit and time limit parameter as

100 units. Please note that the new added parameters are selected in order to not affect

the optimal solution of original problem. The presented model and proposed approach

could easily obtain the optimal solution value (88) for the case of Min (1989) under 1

second (Halse 1992). The solutions are turned out to be computationally demanding

when problem sizes of the instances are increased in the mathematical model. Therefore

the model is not intended for solving large sized problem instances.

The performance of the proposed algorithm is also tested using benchmark instances

for the VRPSPDTL from Salhi and Nagy (1999) based on Christofides et al. (1979).

This problem set includes 14 problem instances in which client numbers vary between

50 and 199. Salhi and Nagy (1999) manipulated 7 original VRP benchmark problem

instances of Christoﬁdes et al. (1979) by imposing a maximum time restriction for the

vehicles, giving a predefined service time, and splitting the original demand between

pickup and delivery loads. The remaining 7 instances were obtained by switching these

pickup and delivery loads.

7.1.2 Numerical results

In order to solve benchmark instances, we firstly performed an extensive experimental

study on the savings heuristic considering different combinations of parameter values:

λ = (1, 0.1, 5); μ = (0, 0.1, 3); ν = (0, 0.1, 2). In the experimental design it is observed

that there is a high interrelation between the savings parameter values and other parame-

ters (vehicle capacity, total duration and service time) of VRPSPDTL problems. As a

result, the following settings were found to provide most reasonable initial solutions to

general subsets of VRPSPDTL problems: λ = 3.5, μ = 1.6, ν = 1.0.

7. Numerical Investigation 98

As a part of preliminary studies, experiments on the sequence of the shaking opera-

tors of the VNS algorithm were conducted in order to determine the most effective se-

quence of the local neighborhood search set. The results demonstrated the effectiveness

of [N1: 3-opt, N2: Swap, N3: Insertion, N4: 2-opt, N5: Exchange (m,n), N6: Cross, N7:

Shift (0,1), N8: Replace (1,1)] sequence. The same sequence is used in the local search

(VND) part of the VNS algorithm. Therefore, k-max and m-max parameters of the VNS

algorithm were set to 8 in the experiments. The total route duration violation acceptance

limit is determined as



= 3 and the vehicle capacity violation acceptance limit is de-

termined as



= 1.2 in these experiments.

In addition to construction heuristic parameters, VNS parameters and route violation

parameters proposed ANS has two major parameters effect the quality of solutions.

These are the APM perturbation counter (p-max) and the ANS termination counter (s-

max). In the proposed ANS approach, the perturbation mechanism is executed after p-

max iterations counted from the last accepted move. The ANS algorithm is terminated

after s-max iterations counted from the last accepted move. In order to determine opti-

mal parameters, an experimental study was conducted with the CMT6X benchmark

problem instance of Salhi and Nagy (1999) with service time (Table 7.1).

Table 7.1: Sensitivity analysis results for algorithm parameters

p-max

5#N

2#N

1#N

0.5#N

s-max

Best

Avg.

Best

Avg.

Best

Avg.

Best

Avg.

1000#N

555.43

556.82

24.1

555.43

556.47

112.1

555.43

47.1

555.43

556.06

32.7

500#N

556.06

557.17

24.1

555.43

556.47

112.1

555.43

47.0

555.43

556.06

32.8

250#N

556.06

557.46

24.2

556.06

556.72

64.5

555.43

556.06

40.1

555.43

556.06

32.7

100#N

556.68

558.03

1.5

556.06

557.29

53.9

555.43

556.31

30.6

555.43

556.43

28.1

50#N

556.68

558.57

0.5

556.06

557.58

14.4

555.43

556.78

18.4

555.43

556.48

24.1

#N: number of clients (50); Best: best solution in 10 replications; Avg.: average solution in 10 replication; T: average best so-

lution time in 10 replications

Table 7.1 implies the importance of perturbation mechanism on approaching to op-

timal solution. As it could be seen in the sensitivity analysis, less perturbation mecha-

nism calling models are working faster but approaching slowly to best known solution

and the solutions are non-robust in ten replications. On the other hand, too much pertur-

bation calling models are working a bit slower, approaching fast but the solutions are

non-robust. Balanced called models (number of client times) are working slower, ap-

proaching to fast, and the solutions are more robust. Hence, the optimal parameter com-

bination for p-max and s-max is 1 (1*50) and 25000 (500*50). Figure 7.1 shows the

improvement of the solution with the iteration counter and improvement of the solution

7. Numerical Investigation 99

with the improvement counter for best solution of CMT13X with service time. Figure

7.1 also implies how perturbation mechanism improves the solution during the itera-

tions by taking from the local optima. The positive effect of the neighborhood structures

are also shown in Table 7.2.

0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2

x 104

1500

1600

1700

1800

1900

2000

2100

2200

2300

050 100 150 200 250 300 350

1500

1600

1700

1800

1900

2000

2100

2200

2300

Figure 7.1: Improvement of the solution during the iterations (CMT13X with service

time)

Table 7.2: The effect of the neighborhood structures (CMT13X with service time)

ANS local search structures

Pure

Sum

ANS Shaking Struc-

tures

197

Sum

317

The entities in Table 7.2 show the positive improvement number of neighborhood

structures in shaking and local search parts. When the interactions between shaking and

local search neighborhood structures are analyzed, the most effective neighborhood

combination is N1&N4 and N1&N5. While intra route structures are more effective in

early stages, the inter-route structures are more effective in later stages of the proposed

ANS. Indeed, the combinations of shaking with intra route structures and local search

with intra route structures are more effective in early stages of the proposed ANS. How-

ever, though the combinations with inter route structures show a smaller quantitative

effect, they play critical roles on route structures. On the other hand, while double per-

turbation structures are effective in early stages, the triple perturbation structures are

more effective in later stages of the solution process.

7. Numerical Investigation 100

The proposed ANS heuristic is first compared with the best solutions of Cluster In-

sertion Heuristics (CIH) by Salhi and Nagy (1999), Insertion Based Heuristics (IBH) by

Detholff (2001), Alternating Heuristic Algorithms (ALT) by Nagy and Salhi (2005),

Large Neighborhood Search (LNS) by Ropke and Pisinger (2006), Tabu Search (TS) by

Montane and Galvao (2006), Reactive Tabu Search (RTS) by Wassan et al. (2008), Iter-

ated Local Search (ILS) by Subramanian and Cabral (2008), Ant Colony System (ACS)

by Gajpal and Abad (2009), Particle Swarm Optimization (PSO) by Ai and

Kachitvichyanukul (2009), Saving Based Ant Algorithm (SBAA) by Catay (2010), and

Nearest Sweep with Perturbation (NSP) by Jun and Kim (2012) for benchmark problem

instances with service time of Salhi and Nagy (1999). The detailed results of the com-

parison of all seven approaches are given in Table 7.3. The proposed ANS algorithm is

run ten times with the same seed sets for each parameter combination in order to meas-

ure their effectiveness and robustness. The best solutions of the problem types are high-

lighted using bold type.

Table 7.3: Computational results for the benchmark problem instances with service

time

Best known

ANS

CMT

#N*

Ref.

BKS

Best

Avg.

Gap%

ALT, ILS, ACS, NSP

555.43

47.0

555.43

0.00

ACS

900.12

901.22

70.3

901.22

0.12

100

LNS, ILS, ACS,

SBAA, NSP

865.50

224.6

865.50

0.00

150

NSP

1161.37

484.0

1162.84

0.00

10X

200

ACS

1386.29

1388.02

1168.8

1390.52

0.12

13X

120

ACS

1542.86

332.7

1543.17

0.00

14X

100

ILS, ACS, SBAA, NSP

821.75

228.5

821.75

0.00

ALT, ACS, NSP

555.43

47.3

555.43

0.00

ACS

900.54

901.22

69.8

901.22

0.08

100

ILS, ACS, SBAA, NSP

865.50

162.7

865.50

0.00

150

NSP

1161.37

527.7

1162.58

0.00

10Y

200

NSP

1392.36

1390.92

1097.4

1391.95

-0.10

13Y

120

ILS, ACS

1542.86

375.3

1542.86

0.00

14Y

100

ILS, NSP

821.75

204.6

821.75

0.00

Avg.

1033.80

1033.94

360.05

1034.50

0.01

* #N: Number of clients; Ref.: Best solution reference; #v: Number of routes; BKS: Best known solution; Best: Best solution in 10

replications; T: Corresponding CPU time; Avg.: Average solution of replications; %Gap: Percentage difference between the best

known and ANS; Avg.: Average of 14 instances

Among 14 instances with service time, the ANS approach could generate a new best

solution for the CMT10Y problem instance. In addition, ANS reproduces best-known

solutions for 10 instances. For the remaining three instances, the gap between the results

of the ANS and the best-known solution is just around 0.10%.

7. Numerical Investigation 101

Secondly, the proposed ANS heuristic is compared with studies excluding service

time. Comparisons with the best solutions of TS by Montane and Galvao (2006), RTS

by Wassan et al. (2008) and ILS by Subramanian and Cabral (2008) are listed in Table

7.4. The best solutions of the problem types are highlighted using bold type.

Table 7.4: Computational results for the benchmark problem instances without service

time

Best known

ANS

CMT

#N*

Ref.

BKS

Best

Avg.

Gap%

ILS

466.77

20.16

466.77

0.00

RTS

663.95

668.35

60.08

668.91

0.66

100

720

720.32

58.90

720.55

0.00

150

ILS

855.74

855.54

120.46

855.70

-0.02

10X

200

ILS

1037.37

1042.12

667.91

1043.14

0.46

13X

120

ILS

846.85

816.87

44.41

818.22

-3.67

14X

100

RTS

644.70

663.50

49.38

663.50

2.83

ILS

466.77

12.67

466.77

0.00

RTS

662.50

664.40

61.07

664.40

0.29

100

TS, ILS

721.40

721.10

47.73

721.12

-0.04

150

ILS

856.74

855.54

180.57

855.68

-0.14

10Y

200

ILS

1036.59

1041.12

355.77

1042.55

0.44

13Y

120

ILS

848.45

809.18

54.40

810.12

-4.85

14Y

100

RTS

659.52

662.22

49.13

662.50

0.41

Avg.

749.09

746.7

127.33

747.13

-0.32

* #N: Number of clients; Ref.: Best solution reference; #v: Number of routes; BKS: Best known solution; Best: Best solution in 10

replications; T: Corresponding CPU time; Avg.: Average solution of replications; %Gap: Percentage difference between the best

known and ANS; Avg.: Average of 14 instances

Among 14 instances without service time, the ANS approach could generate new

best solutions for five problem instances, namely CMT9X, CMT13X, CMT8Y,

CMT9Y and CMT13Y. In addition, ANS reproduces best-known solutions for three

instances. For the remaining six instances, the maximum gap between the results of the

ANS and the best-known solution to the NSP is around 2.83%.

Table 7.5 shows the comparison of approaches for benchmark instances with and

without service time. According to these results, ANS shows the best average solution

(1033.94) for instances with service time. ANS also provided outstanding average solu-

tions (746.70) for instances without service time. Moreover, ANS has also lowest gap

compared with the average of best known solutions for instances with/out service time

within all solution approaches. Generally it is difficult to compare CPU times since dif-

ferent approaches are tested on different computers. In order to make a fair comparison

of execution times, the computers which are used for all approaches are compared with

the help of Passmark Performance Test 7.0 software. Since there are too much factors

effecting on CPU times, approximate equivalent computers are used in the benchmark

test. See the Appendix (Table A.2) for computer benchmark results.

7. Numerical Investigation 102

The average of the computation time for ANS approach is around 6 minutes in the

benchmark problem instance of Salhi and Nagy (1999). In the comprehensive compari-

son with the VRPSPDTL approaches, the scaled average solution time of ANS tolerable

worse than others. Despite the effectiveness of repetitive perturbation mechanism, it

took fairly more computation time (more than 80%) within ANS approach. We note

here that recent effective approaches haven’t provided their optimal computation times.

Concerning robustness, ANS is the most robust of average solution provided approach-

es, since ANS gives lowest average for average solutions for all instances. The variance

among the average of best results and the average of average results is less than the ones

in the other approaches.

Table 7.5: Comparisons of approaches for benchmark instances

With service time

Without service time

Name

Avgb

Avga

Gap%

Avgb

Avga

Gap%

CIH*

1138.50

4.9

0.05

0.07821

IBH

1113.43

0.07714

ALT

1053.36

2.3

0.05

0.01903

LNS**

1093.85

1115.38

519.77

77.97

0.04176

781.86

19.17

6.13

0.05082

RTS

1069.78

1074.80

34.41

3.10

0.03491

763.12

97.29

8.76

0.02563

ILS

1038.08

1050.83

53.42

37.40

0.00425

754.20

767.99

72.02

50.41

0.01364

ACS

1034.01

1035.65

205.48

63.70

0.00031

PSO

1065.86

1085.78

0.03112

SBAA

1042.20

1049.52

0.00823

NSP

1034.95

0.00122

ANS

1033.94

1034.50

360.05

0.00024

746.70

747.13

127.33

0.00356

Best***

1033.69

0.00000

744.05

0.00000

*CIH did not provide solutions for 7X and 7Y; **LNS did not provide solutions for 14Y; ***Average of best known solutions for

all instances. Avgb.: Average of best solutions found in benchmark instances; Avga.: Average of average solutions found in bench-

mark instances; To: Average of best solution time; Tm: In order to make a fair a comparison between algorithms, original solution

times of all approaches are modified according to Table A.2 in Appendix. %Gap: Percentage difference between the best known

solution and methods.

The general conclusion that can be drawn from Table 7.3, Table 7.4 and Table 7.5 is

that the ANS algorithm produces adequate and robust solutions in reasonable time for

the benchmark problems of Salhi and Nagy (1999). The provided best solutions for

CMT10Y with service time and CMT13Y without service time are given in the Appen-

dix (Table A.3 and Table A.4).

7.1.3 Concluding remarks

In this section, we proposed a novel hybrid search method called Adaptive Neighbor-

hood Search (ANS) algorithm based on the Savings Algorithm (SA), Variable Neigh-

borhood Search (VNS) and the Adaptive Perturbation Mechanism (APM) to solve the

7. Numerical Investigation 103

vehicle routing problem with simultaneous pick-up and delivery with time limit

(VRPSPDTL). We used eight local neighborhood search structures as shaking and local

search operators of the VNS algorithm. A Variable Neighborhood Descent (VND) pro-

cedure is used to perform the local search. We use five adaptive perturbation structures

in order to escape from local optima. From the numerical results it can be concluded

that the proposed ANS algorithm generates efficient and robust solutions compared to

existing solution methods for the VRPSPDTL. For 19 out of the 28 benchmark instanc-

es with and without service time from Salhi and Nagy (1999), the ANS algorithm could

obtain new best solutions or reach the best known solution. The main features of the

proposed ANS algorithm are specifically designed sub-procedures as part of the con-

struction heuristic, improvement algorithm and perturbation mechanism to cover the

total vehicle duration limit which is not included in the pure VRPSPD solution methods.

7.2 FCRP application

The presented mathematical model in Section 5.2 has been solved with the proposed

metaheuristic approach in Section 6.1 coded using Matlab R2009a and executed by us-

ing Visual C# 2010 on an Intel Core-2-Duo T5750 2.0 GHz processor with 3 GB RAM.

The performance of the proposed algorithm is tested using a real case study from Aege-

an Islands developed by Sambracos et al. (2004) and generalized by Karlaftis et al.

(2009).

7.2.1 Case study

In this case study, the problem is routing of freight vessels from the port of Piraeus to a

set of 25 islands in the Aegean Sea. See details of the problem at Table A.5 in the Ap-

pendix. The homogenous vessel capacity is 100 small containers, average ship speed is

12 knots, the total delivery amount equals to 464, and the total pickup amount is 235.

The service time to vessels is different in each island and time dead line for vessel is

used as 40 hours. The authors used time limit as soft deadline for supplying islands with

goods. The authors also used a tolerance for approaching an island later than this time

deadline. Such routes are penalized with 5% of the delays.

7. Numerical Investigation 104

7.2.2 Numerical results

In order to solve the defined case study, we performed an extensive experimental design

for savings heuristic considering different combinations of parameter values:

λ = (1: 0.1: 5); μ = (0: 0.1: 3); ν = (0: 0.1: 2). In the experimental design it is observed

that there is a high relationship between the parameter values savings and parameters

(vehicle capacity, total duration and service time) of FCRP. Therefore, following setting

is observed to provide ideal initial solutions on general subset of FCRP:

λ = 3.5; μ = 1.6; ν = 1.0.

As a part of preliminary studies, experiments on the sequence of the shaking opera-

tors of the EVNS algorithm were conducted in order to determine the most effective

sequence of the local neighborhood search set. The results demonstrated the effective-

ness of [N1: 3-opt, N2: Swap, N3: Insertion, N4: 2-opt, N5: Exchange (m,n), N6: Cross,

N7: Shift (0,1), N8: Replace (1,1)] sequence. The same sequence is used in the local

search (VND) part of the EVNS algorithm. Therefore, kmax and mmax parameters of the

EVNS algorithm set to 8 in the experiments. The vehicle capacity violation acceptance

limit is determined as α = 1.2 in these experiments.

In addition to construction heuristic parameters, EVNS parameters, and route viola-

tion parameters, proposed ANS have two algorithm parameters effecting on the quality

of solutions. These are perturbation counter (p-max) and ANS termination counter (s-

max). In the proposed ANS approach, the perturbation mechanism is called after p-max

iterations counted from the last accepted move. The ANS algorithm is terminated after

s-max iterations counted from the last accepted move. A sensitivity analysis performed

in order to determine optimal algorithm parameters. The proposed ANS algorithm is run

ten times with same seed sets for each parameter combination in order to measure their

effectiveness and robustness. Table 7.6 shows that optimal (fast and robust) parameter

combination for p-max and s-max is 25 (≈1*25) and 2500 (=100*25).

The proposed ANS approach is also compared with the best solutions of the GA by

Karlaftis et al. (2009) for the Aegean Islands case study. The case study is also solved

with hard deadline which do not allow delays for approaching an island. In order to

make fair comparison, the case study with both soft and hard deadline is also solved

with EVNS (without perturbation mechanism). Moreover, the GA developed by

Karlaftis et al. (2009) are reprogrammed and validated in order to solve the case study

7. Numerical Investigation 105

with hard deadline (GA-2). The detailed results of the comparison of algorithms are

given in Table 7.7 and best set of routes are shown in Figure 7.2. Please see Table A.6

and Table A.7 in the Appendix for details of the best solutions.

Table 7.6: Sensitivity analysis results for algorithm parameters

p-max

10#N

4#N

2#N

1#N

s-max

Best

Avg.

Best

Avg.

Best

Avg.

Best

Avg.

1000#N

253.96

253.97

147.77

253.96

98.16

253.96

116.50

253.96

33.03

500#N

253.96

253.98

77.80

253.96

98.16

253.96

116.50

253.96

33.03

250#N

253.96

253.98

47.99

253.96

253.97

64.51

253.96

253.97

75.74

253.96

33.03

100#N

254.00

19.88

253.96

254.21

14.38

253.96

254.44

10.06

253.96

253.97

23.36

50#N

254.00

254.67

6.47

253.96

254.22

6.70

253.96

254.44

11.27

253.96

253.98

19.32

p-max

0.5#N

0.25#N

0.125#N

1000#N

253.96

116.86

253.96

179.87

253.96

254.05

139.66

254.00

254.67

2.02

500#N

253.96

254.05

72.72

253.96

253.97

129.47

253.96

254.06

89.05

254.00

254.67

2.02

250#N

253.96

254.07

21.53

253.96

254.06

39.36

253.96

254.07

53.85

254.00

254.67

2.02

100#N

253.96

254.07

19.15

253.96

254.07

25.82

254.00

254.08

25.36

254.00

254.67

2.02

50#N

253.96

254.21

10.72

253.96

254.22

7.29

254.00

254.97

8.76

254.00

254.67

2.02

#N: number of clients (25 islands); Best: best solution in 10 replications; Avg.: average solution in 10 replication; T: average best

solution finding time in 10 replications

Table 7.7: Comparisons of algorithms

Algorithms

Dead line

Number of

routes

Best Fitness

Function value1

Average solu-

tion value2

Avg.Best solu-

tion time (s)3

Soft

260.22

97.54,5

GA-2

Hard

264.00

265.39

33.68

EVNS

Soft

254.00

254.67

2.02

EVNS

Hard

258.12

259.48

1.12

ANS

Soft

253.96

253.97

23.36

ANS

Hard

256.00

256.07

5.57

1 best solution in 10 replication; 2 average of best solutions in 10 replication; 3 average of best solution finding time in 10 replication;

4 best solution of in 1 replication; 5 “Intel Core 2 Duo T5750 2.0 GHz processor with 2 Gb RAM” is around 4.4 times faster than

Karlaftis et al. (2009)’s “Intel Pentium 4 2.53 GHz processor with 512 Mb RAM” according to Passmark Performance Test 7.0

software. Therefore, in order to make a fair a comparison between algorithms, original solution time of Karlaftis et al. (2009) is

modified.

020 40 60 80 100 120 140

100

120

140 Total Fitness = 253.9642

020 40 60 80 100 120 140

100

120

140 Total Fitness = 256.0000

Figure 7.2: Best solution networks for soft and hard time deadline

The proposed ANS algorithm could find the new best solution for the Aegean Is-

lands case study for both soft and hard deadline restriction. On the other hand, EVNS

approach could get results faster solutions; however it is stuck in local optima. The solu-

7. Numerical Investigation 106

tions provided for hard deadline case are more robust and faster than soft dead line case,

since hard line case does not allow time violation for serving islands.

Container transportation capacity of feeder containerships and time deadlines for de-

livering containers to feeder ports are main parameters for FCRP. In addition to sensi-

tivity analysis for algorithm parameters, further sensitivity analysis performed in order

to analyze effect of problem parameters (see Table 7.8).

Table 7.8: Sensitivity analysis results for problem parameters

Deadline

Capacity

Best S.

#Routes

Delay

Best S.

#Routes

Delay

Best S.

#Routes

Delay

Best S.

#Routes

Delay

310.71

2.74

310.03

0.81

309.67

0.01

309.67

0.00

100

255.45

8.46

254.70

5.46

253.96

2.52

253.70

1.45

125

226.38

16.80

225.63

12.33

224.88

8.58

224.31

5.70

150

219.64

17.36

218.89

13.61

218.14

9.86

217.39

6.11

175

205.23

25.96

204.48

20.96

203.73

15.96

202.98

10.96

According to results of sensitivity analysis for problem parameters, as it expected,

fitness value of solution, the number of routes decreases when capacity of barge con-

tainerships and time deadline are simultaneously increased. On the other hand, the de-

crease of fitness value and number of routes increase the average delay on approaching

of containerships to feeder ports. Short delays are considered tolerable in FCRP since

uncertain nature of maritime transportation and feeder ports. However, increase on con-

tainership capacity causes inadmissible delays on approaching since soft time deadline

which directs algorithm to use less containership number. In this context, selection of

barge containership type is purely related to tolerance of time deadlines which is related

to type of transported goods. Therefore, penalty value for penalizing delays on ap-

proaching of containerships to feeder ports is another relevant parameter. Penalty value

is used low for more perishable product transportation and high for less perishable

product transportation. Figure 7.3 shows effect of penalty parameter on solution of the

problem (Capacity 100 TEU and time deadline 40 hour).

According to results of penalty parameter analyze, as expected, fitness value of solu-

tion is increased and average delay is decreased when penalizing value (%) is increased.

While problem is less sensitive to time deadline in lover penalizing value, it is more

sensitive to time deadline in higher values. After 500% penalizing value, problem is

starting to use deadline as hard time deadline for approaching of containerships to feed-

er ports. However, in our experience, total number of routes is not affected by increase

of penalizing value, since current time deadline is adequate for feeder containerships.

7. Numerical Investigation 107

Figure 7.3: Sensitivity analysis results for penalty parameter

7.2.3 Concluding remarks

In this section, we proposed a novel hybrid search method called adaptive neighborhood

search (ANS) which uses the savings algorithm, enhanced variable neighborhood search

and perturbation mechanism in order to solve to the feeder containership routing prob-

lem (FCRP). We used eight local neighborhood search structures as shaking and local

search operators of the algorithm. The proposed approach is tested on a case study from

Aegean Islands and solutions are improved around %3. Moreover, a range of scenarios

and parameters values used in order to test the robustness of the approach through sensi-

tivity analysis. From the numerical results it can be concluded that the proposed ANS

algorithm generates efficient and robust solutions for the FCRP.

7.3 FNDP application

The FND problem presented in Section 5.3 is a highly complex combinatorial optimiza-

tion problem and thus hard to solve by use of standard optimization software. Exact

methods for solving the FND problem are generally not practical for large instances

because of the problem complexity. We therefore employ an adaptive neighborhood

search (ANS) heuristic which has shown to be very efficient for solving the

VRPSPDTL (see Section 7.1). In this section we address the strategic choice of the hub

port, decisions on the size and composition of the fleet of containerships, and ship rout-

ing and scheduling as an integrated planning problem. We consider the Black Sea re-

gion as an application example to analyze the design problem of container feeder net-

works from the perspective of a feeder shipping company commencing its services from

a newly constructed port.

7. Numerical Investigation 108

7.3.1 Implementation

The presented mathematical model in Section 5.3 has been solved with the proposed

ANS approach in Section 6.1 coded using Matlab R2009a and executed by using Visual

C# 2010 on an Intel Core-2-Duo T5750 2.0 GHz processor with 3 GB RAM.

The candidate networks created by the ANS are evaluated using a fitness function.

Since ANS is originally intended to solve the VRPSPDTL with homogenous vehicles

under the objective to minimize the total travel distance within the network, it is neces-

sary to adjust the fitness function of the ANS. In our implementation of the ANS total

operation costs of all routes for the entire sailing season according to the cost functions

(5.25) and (5.26) of Section 5.3 are used as fitness function. The respective procedure

for calculating the fitness values is summarized in Figure 7.4. In the VRPSPDTL appli-

cation candidate routes are generated with the help of neighborhood structures. In this

step constraints (5.29) - (5.39) of the optimization model are checked in order to achieve

feasible solutions.

procedure: fitness function for ANS

input: candidate network

output: total network costs of candidate network (dNC)

start

for each route (r) in the candidate network

initialize a big number for total route cost (RCr)

for each ship type (s)

if (hub and each feeder port departure and arrival loads are feasible for ship type s on

route r) [Eq. (5.35), (5.36), (5.37)]

calculate voyage cycle time of route r with ship type s specifications [Eq. (5.28)]

if (voyage cycle time is feasible by considering maximum voyage dur. limit) [Eq. (5.38)]

calculate required ship number of ship type s for route r [Eq. (5.27)]

calculate variable costs for route r operated with ship type s [Eq. (5.26)]

calculate fixed costs for route r operated with ship type s [Eq. (5.25)]

calculate total costs of route r operated with ship type s (dRCr)

if (dRCr < RCr )

update RCr with dRCr

end if

end for

calculate total network costs of candidate network (dNC = RC𝑟)

𝑟 [Eq. (5.24)]

end

Figure 7.4: Calculation of the FNDP fitness function

7. Numerical Investigation 109

Apart from the network routes the ANS determines the fleet mix, the number of re-

quired ships according to Equation (5.27) and their deployment to routes in the candi-

date network. Based on these data the total voyage cycle of a ship on a route is achieved

as given by Equation (5.28), i.e. considering the related port service times, travel times

between ports, lay-up times etc. Figure 7.5 shows an example of a route-ship-port

schedule for 30 days of operation with 5 days service frequency, 3 feeder ports in se-

quence, and 3 required ships for a route in the network.

*H.L: loading time at hub port; H-1, 1-2, 2-3, 3-H: port-to-port travel time; 1, 2, 3: feeder port service time including, loading,

unloading and set-up times; H.U: unloading time at hub port; L.U.: lay-up time of ship for next voyage

Figure 7.5: An example of a route-ship-port schedule

7.3.2 Case study

The fact that the considered region is surrounded by several seas – the Black Sea, Medi-

terranean Sea, Adriatic Sea, Ionian Sea, Aegean Sea, and Marmara Sea – makes mari-

time shipping a prime area for sustained growth (see Figure 7.6). Container feeder ship-

ping lines offer crucial transport connections between the hinterland of this region and

global trunk shipping lines. The feeder shipping dynamics of the region are mainly re-

lated to container transportation volumes of the trunk shipping lines between Far East

and Europe. In recent years, parallel to the increase of container transportation volumes

on the global trunk shipping lines, an increase of the total container handling volume is

observed in the regional feeder ports. This is particularly true for ports in the Black Sea

region. Hence, the outlook for the maritime transportation market in the region is very

promising (Varbanova 2011a).

Turkey’s ideal location between Asia and Europe gives its ports a competitive ad-

vantage and opportunity to develop into major transhipment hub ports. However, so far

Turkish ports primarily serve their national needs and remain outside the major trunk

lines (Kulak et al. 2013). This situation results in maritime container transport mainly

executed by feeder lines that serve the Turkish ports from the East Mediterranean hub

7. Numerical Investigation 110

ports. In this regard, Turkey has significant potential for getting stronger involved in

regional maritime transport and consequently several projects for the development of

intermodal transport are being initiated. One of these projects is the construction of a

hub port in Izmir’s Candarli district in order to improve Turkey’s hub port potential in

the East Mediterranean and especially in the Black Sea region. According to the project

plan, the Northern Aegean Candarli port will take its place among the world’s largest

ports after its first part is completed in 2013 and it will be able to handle 12 million tons

of container freight annually in its ultimate configuration. The potential market areas of

Candarli as a hub port can be categorized into four sub-regions: the Black Sea, the

Marmara Sea, the East Mediterranean and the Aegean Sea.

Mersin

Antalya

Candarli

Izmir

Gebze

Ambarli Haydarpasa

Gemlik

Thessaloniki

Piraeus

Limassol

Lattakia

Beirut

Haifa

Ashdod

Port Said

Alexandria

Burgas

Varna

Ilyichevsk Odessa

Novorossiysk

Poti

Trabzon

Aliaga

Aegean Sea

Mediterranean Sea

Black Sea

Sea of Azov

Constantza

TURKEY

Damietta

Batumi

Sea of Marmara

Figure 7.6: Regional feeder and hub ports

A particular feeder liner shipping company currently operates a feeder network with

a hub port at Port Said in Northern Egypt. However, after opening of the new Candarli

port, the company will possibly redesign its current feeder network with Candarli as a

7. Numerical Investigation 111

new hub port. Therefore, in this study three different strategic options for hub ports are

considered.

 The first strategic option corresponds to the current configuration with Port Said

as hub for feeder ports in the Black Sea region. The main advantage of this op-

tion is the closeness to the Suez Canal through which almost all of the Asia-

Europe shipping routes pass.

 In the second strategic option the new Candarli port replaces Port Said as a hub

port for the Black Sea region. This option is based on the assumption that Can-

darli will be a firm part of the global shipping routes.

 The third strategic option is a mixed case in which two hub ports are established.

Namely Port Said serves as a link to the main global shipping lines and at the

same time as regional hub port for the East Mediterranean ports. Candarli will

serve as a second regional hub port for the Black Sea, the Sea of Marmara and

the Aegean Sea ports and with daily direct connections to Port Said via mid-sized

ships.

These strategic options are tested under different time deadline and service fre-

quency conditions for a 52-week sailing season. In this region, the concerned feeder

liner company has 36 contracted container terminals at 26 feeder ports which have a

total daily demand of 3321 TEU and a total daily supply of 2151 TEU on average (see

Table A.8 in the Appendix for details of the ports). Because of the limited berth depth at

some regional ports and well-known traffic bottlenecks at the Bosporus and Dardanelles

straights, ships of three different sizes are considered in the numerical experiment. The

major cost parameters for all ship types are shown in Table 7.9.

7.3.3 Numerical results

In order to provide decision support for the feeder network design problem faced by the

Turkish company, we proceed with our experiments in the following order. First the

strategic options for choosing the hub port are evaluated (Section 7.3.3.1). Second, the

impact of different scenarios for the long-term development of transportation volume in

the Black Sea region is analyzed (Section 7.3.3.2).

7. Numerical Investigation 112

Table 7.9: Parameter values for ship types

Parameter

Unit

Ship1

Ship 2

Ship 3

Capacity

TEU

4300

2600

1200

Operating speed

(knots)

22.60

19.90

17.40

Fuel consumption (on sea)

(tons/hour)

5.26

2.82

1.51

IFO 180 price (on sea)

($/ton)

647.50

Fuel consumption (at port)

(tons/hour)

0.26

0.14

0.08

MGO price (at port)

($/ton)

890.00

Chartering cost

($/day)

12772.00

7579.00

5866.00

Operating costs

($/day)

11520.00

8887.00

6023.00

Port charges

($/call)

35000.00

29000.00

22000.00

Lay-up time

(hour/call)

28.80

24.00

16.80

Set-up time

(hour/port)

2.00

1.80

1.50

Planning period

Days

364

Sources: Stopford (2009), VHSS (2013), BunkerIndex (2012)

7.3.3.1.Strategic options for feeder networks

The three basic strategic options to be considered are the locations of hub port in

Port Said and Candarli, respectively, and a combined network design with these two

transhipment hubs connected by a shuttle service of feeder ships. In the experiments

two additional network design parameters are evaluated. As for shipping frequencies we

compare the departure of services every 7 or 3.5 days, respectively. For each frequency

the time deadline for voyages is varied between 3 and 4.5 weeks.

The specific research issues addressed in our numerical investigation are the follow-

ing:

 Does the time deadline imposed on the voyages represent a major factor in the

design of the network configuration?

 How does the voyage frequency impact the cost performance of the various net-

work configurations?

 Which of the three strategic options for hub ports would be favourable in terms

total yearly costs?

The perturbation mechanism is called after

1 feeder portnumber

, i.e. 36, iterations

counted from the last accepted move. The total route duration and the vehicle capacity

violation acceptance limit (α1 and α2) are used as 10%. This rule aims to allow custom-

ers to join another route for possible future improvements. The termination condition of

the ANS algorithm is used as maximum number of iterations between two improve-

ments of the best solution. The termination condition is set to

100* feeder portnumber

7. Numerical Investigation 113

iterations without improvement. The proposed ANS algorithm is run ten times with dif-

ferent random seeds in order to measure its robustness.

Table 7.10 shows the total costs of the current and alternate hub port options under

various time deadline and service frequencies. Total costs include chartering costs, op-

erating costs, administration costs, on-sea bunker costs, on-port bunker cost and port

charges for a 52-week sailing season. Computational times depending on the structure

of the feeder network varied between 10 and 60 seconds.

Table 7.10: Scenario results for alternative hub port locations

Scenario

Hub

Frequency

(days)

Deadline

(days)

Minimum total

cost (x1000)

Average total cost

(x1000)

Average CPU

time

Port Said

3x7

286548.47

286978.61

56.38

Port Said

3.5x7

286911.48

287691.02

58.60

Port Said

4x7

286420.04

287616.77

39.80

Port Said

4.5x7

287388.27

287893.72

51.95

Port Said

3.5

3x7

339726.79

342734.19

25.03

Port Said

3.5

3.5x7

339726.79

341642.78

31.06

Port Said

3.5

4x7

339726.79

341565.97

23.73

Port Said

3.5

4.5x7

339726.90

342141.54

18.43

Candarli

3x7

257526.74

258052.94

36.43

Candarli

3.5x7

255733.20

257470.14

50.02

Candarli

4x7

255341.87

257051.03

30.97

Candarli

4.5x7

255341.87

257647.93

38.38

Candarli

3.5

3x7

296789.81

298547.32

42.07

Candarli

3.5

3.5x7

298157.27

300052.71

20.21

Candarli

3.5

4x7

296831.31

299795.88

17.26

Candarli

3.5

4.5x7

296789.81

297920.09

29.81

Mixed

3x7

368529.88

369673.02

17.07

Mixed

3.5x7

367462.31

369468.74

17.64

Mixed

4x7

367099.69

369206.60

23.37

Mixed

4.5x7

367462.31

368277.30

30.25

Mixed

3.5

3x7

403355.76

405330.18

10.36

Mixed

3.5

3.5x7

401789.60

402769.03

14.50

Mixed

3.5

4x7

401789.60

401818.82

15.93

Mixed

3.5

4.5x7

401789.60

402453.47

12.12

The first conclusion that can be drawn from the results displayed in Table 7.10 is

that the effect of the time deadline is practically negligible. Even the largest deviation

observed for Candarli and the 7-days frequency options (no. 9-12) are less than 1%.

However voyage frequencies have a major impact on the cost performance. Reduc-

ing the voyage frequency for Port Said from 7 to 3.5 days causes a cost increase of

18.65%. Respective values are 16.2% for Candarli and 9.44% for the configuration with

two hubs.

The main research question addresses the choice of the hub location. It can be seen

from the results shown in Table 7.10 that the mixed hub option causes total costs of

7. Numerical Investigation 114

$367,099,690 (option no. 19 with 7-day service frequency and 4 weeks deadline) and

thus is clearly outperformed by the single-hub configurations. This cost disadvantage is

mainly due to the additional transhipment operations at Candarli. As for the single-hub

configurations the existing hub port option of Port Said shows minimum total costs of

$286,420,040 (option no. 3 with 7-day service frequency and 4 weeks deadline) while

the projected hub port of Candarli achieves minimum total cost of $255,341,870 (option

no. 11 with 7-day service frequency and 4 weeks deadline). Considering only network-

wide cost figures the Candarli option would allow cost savings of 12.2% compared to

the existing feeder network configuration with Port Said as hub. The resulting feeder

routes for Port Said (option no. 3) and Candarli (option no. 11) are shown in Figure 7.7.

020 40 60 80 100 120 140

100

120

140 Total Distance = 286420.0375

020 40 60 80 100 120 140

100

120

140 Total Distance = 255341.8666

Figure 7.7: Feeder route networks for Port Said (left) and Candarli port (right)

Table 7.11 presents a comparison of costs, fleet and voyage characteristics of the

two single-hub configurations. As in global trunk lines, feeder shipment is highly sensi-

tive to bunker fuel costs as they represent 26.67% (Port Said) and 20.83% (Candarli) of

total costs. However, these shares are significantly lower compared to global trunk lines

due to the density of the network and the relatively short transportation distances. In

turn feeder networks show a higher share of ship based fixed costs such as chartering,

operating and port charges. Since Candarli has shorter distances to regional feeder ports,

relatively small containerships are employed. In contrast, the Port Said based feeder

network utilizes slightly more mid-sized containerships. Large-sized feeder ships of

4300 TEU are not appropriate for both hub alternatives because of the relatively high

fixed costs. It could be expected, however, that in case the network dimension is

enlarged and total demand increases, larger ships will become more attractive in order

to meet the balance between fixed and variable costs. It is also shown in Table 7.11 that

total voyage durations of 297.23 hours are slightly lower for the Port Said option com-

7. Numerical Investigation 115

pared to Candarli with 308 hours. In both cases the major share of the voyage durations

of more than 60% occurs for the stay in the hub and in the feeder ports. As expected the

on-sea voyage duration is lower for the Candarli option due to its geographical location

closer to the Black Sea region. The best solution achieved for Candarli is given in detail

in Table A.9 in the Appendix.

Table 7.11: Feeder network comparison of the Port Said and Candarli port options

Parameter

Port Said

Candarli

Costs

Total costs (´000 $)

286,420.04

255,341.87

Chartering costs

20.80%

22.30%

Operating costs

23.62%

25.42%

Bunker costs (on sea)

26.67%

20.83%

Bunker costs (at port)

4.83%

5.38%

Port charges

24.09%

26.07%

Fleet

Number of routes

Total number of ships

1200 TEU

20.44%

27.27%

2600 TEU

79.56%

72.73%

4300 TEU

0.00%

Voyages

Total avg. duration (Hour)

297.23

308.00

On sea

23.71%

17.69%

In feeder ports

41.31%

43.17%

In hub port

20.39%

21.27%

Lay-up times

6.96%

6.82%

Idle times

7.64%

11.06%

A specific drawback of the Candarli option compared to Port Said is certainly its lo-

cation of about 220 nautical miles farther away from the main global shipping lines.

Under one daily East-Westbound and West-Eastbound service assumption, the extra

costs for operating this transhipment service would almost compensate the saving in

operational costs.

7.3.3.2.Demand scenarios

For the future development of the feeder network the expected growth of the transporta-

tion market in the Black Sea region is an essential factor. According to forecasting re-

ports of Ocean Shipping Consultants (2011), container handling demand in the region

will continue to increase yearly by 25% till 2025. Therefore, a sensitivity analysis is

performed to assess the influence of this factor on the cost performance of the Port Said

and Candarli network configurations. Based on this expectation for four subregions, 16

different market scenarios are created in order to evaluate the network costs for ex-

changing the current hub port. Scenario 1 corresponds to the current market situation. In

the further scenarios combinations of market volume increase in one, two and three re-

7. Numerical Investigation 116

gions, respectively, are assumed. Finally, scenario 16 corresponds to a 25% market vol-

ume increase in all four regions.

Results of the sensitivity analysis summarized in Table 7.12 show network costs for

the two candidate hub ports under equivalent demand increase assumptions. According

to the results of the sensitivity analysis, Candarli outperforms Port Said in all demand

scenarios because of its advantageous geographical position. Candarli's superiority,

however, is considerably smaller when only a market volume increase in the East Medi-

terranean and the Aegean Sea region is assumed. Otherwise, Candarli benefits from

increased market volumes in the Black Sea and the Sea of Marmara region.

Table 7.12: Sensitivity analysis of market volume increase

Assumed market volume increase*

Total costs for alternative hub ports**

Scenario

no.

Black Sea

region

Sea of

Marmara

region

Aegean

Sea region

East Med.

sea region

Port Said

(´000 $)

Candarli

(´000 $)

Difference

(´000 $)

286,420.04

255,341.87

31,078.17

309,076.85

269,818.90

39,257.95

298,126.92

268,435.94

29,690.98

295,888.30

264,097.58

31,790.72

298,868.95

273,022.57

25,846.38

323,199.03

283,684.09

39,514.94

310,919.54

275,603.45

35,316.09

306,960.82

280,852.65

26,108.17

318,259.61

279,308.36

38,951.25

320,500.69

288,631.05

31,869.64

312,501.39

283,657.26

28,844.13

332,576.84

291,443.00

41,133.84

322,131.10

291,084.21

31,046.89

334,665.42

299,057.31

35,608.11

333,242.55

295,495.20

37,747.35

344,605.35

308,725.32

35,880.03

*o indicates that the company will maintain its current market share; + indicates that the company will increase its current market

share in the region by 25%. ** The results show the best of 10 replications of the heuristic for alternative hub ports with 4 weeks

deadline and 7 days service frequency for a 52 week sailing season.

7.3.4 Concluding remarks

In this section, we focus on the potential hub role of a new port (Candarli) in the East

Mediterranean and Black Sea region and apply a heuristic procedure to solve the feeder

network design problem faced by a short-sea shipping company. Based on the container

transportation demand at feeder ports, the feeder network and fleet mix, the composition

of routes and the schedule of the vessels operating on these routes are determined by

minimizing total operational costs. A mathematical model of the feeder network design

problem has been developed. Because of the complexity of the optimization problem an

efficient heuristic solution procedure was applied.

7. Numerical Investigation 117

In the numerical investigation the cost performance of three strategic options for hub

port configurations has been compared. From the numerical results it can be concluded

that Candarli as a new hub port offers significant cost savings compared to Port Said

which is currently used as a hub port by the considered company. However, these cost

savings would be compensated with additional transhipment cost for the Port Said -

Candarli services which are needed to connect Candarli to the global trunk shipping

lines. Therefore, additional factors like service quality and handling efficiency at the

hub ports as well as waiting time in the queue of the hub ports play an important role in

the development of the company's feeder network configuration. Certainly, the new

Candarli port has great market potential as long as port authorities keep container han-

dling costs and service quality at a favourable level.

7.4 LSND under unstable demand environments

The LSND problem presented in Section 5.4 is a highly complex combinatorial optimi-

zation problem and is thus hard to solve with the use of standard optimization software.

Exact methods for solving the LSND problem are generally not practical for large in-

stances because of the problem complexity. We therefore employ a simulation-

optimization based solution framework which contains a hybrid solution heuristic called

Adaptive Neighborhood Search (Section 6.1) and a simulation and artificial neural net-

works based forecasting model (Section 6.2) is proposed to solve the joint problem.

In this section, we consider the Black Sea region as an application example to ana-

lyze the design problem of liner shipping networks under unstable demand environ-

ments from the perspective of a feeder shipping company commencing its services from

a newly constructed port. The design of the service network is revised at the beginning

of every period in response to changes in demand patterns for a season estimated with a

simulation based forecasting framework. Changes to the service network may include

introducing new routes, and schedules as well as fleet deployments which could contain

chartering in new ships or chartering out unnecessary ships. This section also employs

various service scenarios in order to better help decision makers of liner shipping pro-

viders. These scenarios contain different periodical approaches, different demand allo-

cations, different numbers of owned ships at the start of sailing season, and different

ship prices to provide a very high degree of ﬂexibility for planning decisions under un-

stable demand environments.

7. Numerical Investigation 118

7.4.1 Implementation

The mathematical model presented in Section 5.4 has been solved with the proposed

ANS approach in Section 6.1 coded using Matlab R2009a and executed by using Visual

C# 2010 on an Intel Core-2-Duo T5750 2.0 GHz processor with 3 GB RAM.

The candidate networks created by the ANS according to periodical demands esti-

mated by forecasting framework are evaluated using a fitness function. Since ANS is

originally intended to solve the VRPSPDTL with homogenous vehicles under the objec-

tive to minimize the total travel distance with stable demands in the network, it is neces-

sary to adjust the fitness function of the ANS according to the unstable demand envi-

ronment of the LSND problem. In our implementation of the ANS total operation costs

of all routes and all periods for the entire sailing season according to the cost functions

(5.41) and (5.42) of Section 5.4 are used as the fitness function. The respective proce-

dure for calculating the fitness values is summarized in Figure 7.8. In the VRPSPDTL

application candidate routes are generated with the help of neighborhood structures. In

this step constraints (5.45) - (5.57) of the optimization model are checked in order to

achieve feasible solutions. The regret value represents the difference between the ship

types and deployments. The assignment with the highest regret value is assigned to the

related route by considering on hand ship numbers. After assigning on hand ships to

routes, the remaining empty routes are operated with charter ships. The idle on hand

ships are chartered out to the market.

Apart from the network routes the ANS determines the fleet mix, the number of re-

quired ships according to Equation (5.43) and their deployment to routes in the candi-

date network under unstable demand environment for each period. Based on these data

the total voyage cycle of a ship on a route is achieved as given by Equation (5.44), i.e.

considering the related port service times, travel times between ports, lay-up times etc.

7. Numerical Investigation 119

procedure: fitness function for ANS

input: candidate network

output: total network costs of candidate network (dNC)

start

for each route (r) in the candidate network

initialize a big number for total route cost (RCr)

for each ship type (s)

if (hub and each feeder port departure and arrival loads are feasible for ship type s on

route r)

calculate voyage cycle time of route r with ship type s specifications

if (voyage cycle time is feasible by considering maximum voyage dur. limit)

calculate required ship number of ship type s for route r

calculate variable costs for route r operated with ship type s

calculate fixed costs for route r operated with chartered ship type s

calculate fixed costs for route r operated with owned ship type s

calculate total costs of route r operated with chartered ship type s (dRCrs1)

calculate total costs of route r operated with owned ship type s (dRCrs2)

end if

end for

create regret value index (variance) matrix between (dRCrs1) and (dRCrs2)

sort rows and columns and rows of the matrix in descending order

assign on hand ships to routes by considering on hand ship number and regret value

charter in necessary ships

charter out unnecessary ships

calculate total network costs of candidate network (dNC)

end

Figure 7.8: Calculation of the LSND fitness function

7.4.2 Case study

The fact that the considered region is surrounded by several seas – the Black Sea, Medi-

terranean Sea, Adriatic Sea, Ionian Sea, Aegean Sea, and Marmara Sea – makes mari-

time shipping a prime area for sustained growth (see Figure 7.10). Container feeder

shipping lines offer crucial transport connections between the hinterland of this region

and global trunk shipping lines. The feeder shipping dynamics of the region are mainly

related to container transportation volumes of the trunk shipping lines between the Far

East and Europe. In recent years, parallel to the increase of container transportation vol-

umes on the global trunk shipping lines, an increase of the total container handling vol-

ume has been observed in the regional feeder ports. This is particularly true for ports in

the Black Sea region. Hence, the outlook for the maritime transportation market in the

region is very promising (Varbanova 2011a).

7. Numerical Investigation 120

Mersin

Antalya

Candarli

Izmir

Gebze

Ambarli Haydarpasa

Gemlik

Thessaloniki

Piraeus

Limassol

Lattakia

Beirut

Haifa

Ashdod

Alexandria

Burgas

Varna

Ilyichevsk Odessa

Novorossiysk

Poti

Trabzon

Aliaga

Aegean Sea

Mediterranean Sea

Black Sea

Sea of Azov

Constantza

TURKEY

Damietta

Batumi

Sea of Marmara

Figure 7.9: Regional ports

In this region, a particular feeder liner shipping provider would like to design its

service feeder network with a new hub port at Candarli in Turkey. Since liner shipping

has been directly affected by financial, political and seasonal conditions, the provider

would like to design its service networks by considering the seasonal demand fluctua-

tions in this region.

The considered problem is tested under a four-week service time deadline and sev-

en-day service frequency conditions for a 52-week sailing season. In this region, the

concerned feeder liner shipping provider has 36 contracted container terminals at 26

feeder ports in 12 countries. Table A.10 in the Appendix shows detailed information

about the terminals, including country and sub-region information, market share of a

shipping line provider in related terminals, operation efficiency of terminals, and yearly

total throughputs of related terminals between 2005 and 2011.

7. Numerical Investigation 121

Because of the limited berth depth at some regional ports and well-known traffic

bottlenecks at the Bosporus and Dardanelles straights, ships of three different sizes are

considered in the numerical experiment. The major cost parameters for all ship types are

shown in Table 7.13.

Table 7.13: Extended parameter values for ship types

Parameter

Unit

Ship1

Ship 2

Ship 3

Capacity

TEU

4300

2600

1200

Operating speed

(knots)

22.60

19.90

17.40

Fuel consumption (on sea)

(tons/hour)

5.26

2.82

1.51

IFO 180 price (on sea)

($/ton)

647.50

Fuel consumption (on port)

(tons/hour)

0.26

0.14

0.08

MGO price (on port)

($/ton)

890.00

Chartering costs (charter in)

($/day)

12772.00

7579.00

5866.00

Amortization costs (on hand)*

($/day)

6386.00

3789.50

2933.00

Rent price (charter out)**

($/day)

9579.00

5684.25

4399.50

Operating costs

($/day)

11520.00

8887.00

6023.00

Port charges

($/call)

35000.00

29000.00

22000.00

Lay-up time

(hour/call)

28.80

24.00

16.80

Set-up time

(hour/port)

2.00

1.80

1.50

Planning period

Days

364

Sources: Stopford (2009), VHSS (2013), BunkerIndex (2012),* Amortization cost used as 50% of charter in cost, ** Charter out

price used as 75% of charter in cost.

7.4.3 Demand estimation

In this region, the statistical databases on container throughput of the ports are scare and

hard to handle seasonal throughputs from the port authorities. For that reason, yearly

throughputs of regional container terminals are firstly decomposed into monthly supply

and demand amounts by using monthly import and export foreign trade rates of related

port countries. Table A.11 in the Appendix shows monthly import and export foreign

trade volumes of related countries between 2005 and 2011. Table 7.14 shows an exam-

ple decomposition of the Odessa container terminal in 2011. The percentages of month-

ly export and import trades in yearly total foreign trade volumes are calculated. Then

the percentages are used in order to find the monthly statistics of related ports.

Secondly, decomposed monthly statistics are used in the proposed ANNs approach

in order to forecast monthly demand and supply throughputs of terminals. Figure 7.10

shows monthly decomposed demand and supply throughputs of the Odessa container

terminal between 2005 and 2011, and monthly forecasted throughputs for 2012.

7. Numerical Investigation 122

Table 7.14: An example of monthly throughputs decomposition

Total

Trade Export

($ Million)*

4621

5379

5382

5603

5969

5889

5365

5769

5974

5716

6283

6459

68409

Trade Import

($ Million)*

5037

6463

7016

6298

6766

6772

6522

7208

7412

7545

7675

7892

82606

Export in for-

eign trade (%)

3.06

3.56

3.71

3.95

3.90

3.55

3.82

3.96

3.79

4.16

4.28

45.3

Import in for-

eign trade (%)

3.34

4.28

4.65

4.17

4.48

4.32

4.77

4.91

5.00

5.08

5.23

54.7

Container

export (TEU)

13883

16160

16169

16833

17933

17693

16118

17332

17948

17173

18876

19405

205523

Container

import (TEU)

15133

19417

21078

18921

20327

20345

19594

21656

22269

22669

23058

23710

248177

*Source: Ukraine’s monthly foreign trade in goods (2011), Total throughputs of the Odessa container terminal is 453700 TEU in

2011.

Figure 7.10: An example of monthly throughputs estimation

In order to reflect fluctuations in daily operations, a Monte Carlo simulation frame-

work is designed according to expert opinions from the port and shipping authorities.

The designed framework contains two sub-modules, which create a final fluctuation

coefficient. For each month, the ordinary day throughputs (forecasted throughput

amount of the month / the number of days in the month) are fluctuated with this coeffi-

cient. In the designed framework, the first module is used to reflect fluctuations within

weekdays and the second module is used to reflect fluctuations within month days. Ta-

ble 7.15 and Table 7.16 show designed simulation coefficients for week and month days

according to expert opinions. In the tables, coefficients represent operation workload

expectations of the days, and workload ratios represent situations of that day compared

to an ordinary day. Table 7.17 shows an example calculation for fluctuation coefficients

of ordinary days.

7. Numerical Investigation 123

Table 7.15: Simulation coefficients for week days

Days

Low (0.8)

Mid (1)

High (1.2)

Monday

20%

40%

Tuesday

30%

60%

10%

Wednesday

40%

20%

Thursday

40%

50%

10%

Friday

10%

30%

60%

Saturday

10%

50%

40%

Sunday

50%

40%

10%

Table 7.16: Simulation coefficients for month days

Days

Low (0.7)

Mid (1)

High (1.3)

First 5 days

10%

50%

40%

Mid-days

40%

20%

Last 5 days

10%

20%

70%

Table 7.17: An example of daily throughputs fluctuation

Date

Day

Random

number

Week

day coef-

ficient

Random

number

Month day

coefficient

Final

fluctuation

coefficient

Ordinary day

demand

Fluctu-

ated

demand

05 October

2011

Wed.

0.00543

0.80

0.85504

1.30

1.04

6405

6661

06 October

2011

Thu.

0.66611

1.00

0.93674

1.30

6405

8327

Thirdly, the proposed fluctuation simulator is run 100 times for each day of each

month by using forecasted throughputs. In addition, diﬀerent random components of the

future demand and supply throughput expectations of terminals are obtained. Figure

7.11 shows average daily fluctuated demand and supply throughputs of the Odessa con-

tainer terminal for a 364-day sailing season in 2012. Since the considered problem is

tested under seven-day service frequency, the daily throughputs are collected into week-

ly throughputs.

Figure 7.11: An example of daily throughputs simulation

Figure 7.12 shows weekly expected demand and supply throughputs of the Odessa

container terminal and total demand and supply throughputs of the region for a 52-week

7. Numerical Investigation 124

sailing season in 2012. These weekly throughput estimations are used in the case study

to design feeder service networks under unstable demand environments.

Figure 7.12: Examples of weekly demand estimations of the Odessa container terminal

(right) and whole region (left)

7.4.4 The impact of seasonal demand fluctuations

In order to analyze how the seasonal demand fluctuations are affecting the service net-

works of shipping lines, the developed ANS algorithm is run ten times with different

random seeds for each week during a 52-week sailing season by using the forecasted

demands. Figure 7.13 shows the region’s minimum total cost, which includes chartering

costs, operating costs, administration costs, on-sea bunker costs, on-port bunker cost

and port charges for a 52-week sailing season. In other words, the figure shows how

seasonality and demand fluctuations on throughputs are affecting the total service cost

of the region. The optimal total cost of the region changes from $238,940,000 to

$320,690,000 meaning that there is a 34.21% cost difference between the 1st and 38th

weeks of the sailing season. Please note that these total cost figures represent total cost

of the entire planning horizon by using related week’s demand.

Figure 7.13: Minimum total cost of the region for a 52-week sailing season

7. Numerical Investigation 125

Since total voyage distances of feeder networks are less than those of trunk net-

works, total network costs contain more ship based fixed costs, such as chartering, op-

erating and administration (Polat et al. 2013). Therefore, the cost difference within

weeks mainly results from the number of service routes, the number of necessary ships

and the types of these ships. Figure 7.14 and Figure 7.15 show how these parameters

change during the season.

In Figure 7.14, the optimal route number of the network changes from 13 to 17.

Since hub port Candarli has shorter distances to feeder ports, small and mid-sized con-

tainerships are employed in low demand seasons; relatively mid-sized containerships

are employed in regular demand seasons; and big ships only start to be employed in

high demand seasons. Consequently, 34.70% of the routes are serviced by small ships,

64.78% by mid-sized ships, and 0.53% by big ships. On the other hand, the total slot

capacity of the routes changes parallel with the total import of the region. Out of these

slots, 19.61% are owned by small ships, 79.32% by mid-sized ships, and 1.07% by big

ships. The necessary slot capacity fluctuates between 25,400 TEU (28.35% by small

and 71.65% by mid-sized ships) and 37,500 TEU (19.20% by small, 69.33% by mid-

sized and 11.47% by big ships). This means that there is a 47.64% necessary slot capac-

ity difference between the 1st and 37th weeks of the sailing season.

Figure 7.14: Necessary number of routes with ship types and slot capacity

In Figure 7.15, the necessary ship size and mix of the service network changes from

23 ships (39.13% by small and 60.87% by mid-sized) to 30 ships (30.00% by small,

63.33% by mid-sized, and 6.67% by big sized). This number determines the minimum

number of ships required to service the region, with seven-day frequency, according to

the related week’s demand expectation for a 52-week sailing season. Therefore, the

number of required ships is more critical than the number of routes for the total cost of

7. Numerical Investigation 126

the network. The slot capacity of ships fluctuates between 47,200 TEU (22.88% by

small, and 72.12% by mid-sized ships) and 71,800 TEU (8.36% by small, 79.67% by

mid-sized, and 5.99% by big ships). This means that there is a 47.64% necessary slot

capacity difference between the 1st and 38th weeks of the sailing season. On the other

hand, the utilization ratio of total slot capacity fluctuates between 68.71% and 80.04%,

which is mainly the result of import - export imbalances (avg. 1.49) in the region.

Figure 7.15: Necessary minimum number of ships with types and capacity utilization

The decisions in tactical planning operations of liner shipping, such as fleet size and

mix, fleet deployment, ship routing and scheduling are made according to container

throughput estimations. However, these throughputs are highly affected by unstable

financial, political and seasonal conditions. In this study, we focus on developing a

Monte Carlo simulation and artificial neural networks based forecasting frame to ana-

lyze the impact of these conditions on the liner shipping feeder service network design.

The proposed model implementation has been tested for the liner shipping feeder ser-

vice in the Black Sea region. The optimal feeder networks are calculated according to

the forecasted throughputs of the region terminals for each week during a 52-week sail-

ing season. The results show that the total cost of the region service network is affected

around 34.21%, the total slot capacity is affected around 47.64%, and utilization of the

slot capacity, the necessary ship type and mix is highly affected by these unstable condi-

tions. These figures show the importance of dynamic and flexible feeder service net-

work designs in liner shipping in making effective and efficient plans.

7.4.5 Experimental design

The previous section assumes that shipping line providers allow weekly service network

design change. However in practice, due to market service restrictions, providers of

7. Numerical Investigation 127

shipping lines could change their service networks only a couple of times within a year-

ly planning horizon. In this section, we therefore evaluate service network designs by

allowing a limited number of changes in the service network. In the content of this

study, we use monthly, bi-monthly, quarterly, trimester, semi-annual, and annual sea-

sonal periods in order to update demand figures and service network designs. Table 7.18

shows these seasons and shows week allocations to these seasons, thus demonstrating

how many weeks each season contains.

Table 7.18: Seasonal week allocations to periods (Scenario A)

Approach

Number of peri-

odical seasons

Jan.

Feb.

Mar.

Apr.

May

Jun.

Jul.

Aug.

Sep.

Oct.

Nov.

Dec.

Since each service week has different demand levels, it is necessary to assign com-

mon supply and demand figures to related seasons. In this study, we compared five dif-

ferent approaches in order to assign supply and demand figures to seasons (Table 7.19).

The first approach assigns each terminal’s maximum demanded (demand or supply)

week figure as seasonal demand for the terminal. The second approach assigns the max-

imum total demanded week’s figures as seasonal demand for each terminal. The third

approach assigns each terminal’s minimum demanded week figure as seasonal demand

for the terminal. The forth approach assigns the minimum total demanded week’s fig-

ures as seasonal demand for each terminal. The fifth approach assigns seasonal demand

according to the average seasonal demand of each terminal in the season.

Table 7.20 shows an example of seasonal demand assignments according to demand

and supply of four ports and four weeks.

Table 7.19: Seasonal demand assignment approaches (Scenario B)

Approach

According to maximum demanded week of each terminal in the season

According to maximum total demanded week of the season

According to minimum demanded week of each terminal in the season

According to minimum total demanded week of the season

According to average seasonal demand of each terminal in the season

7. Numerical Investigation 128

Table 7.20: Example of seasonal demand assignments

Terminal A

Terminal B

Terminal C

Terminal D

Demand

Supply

Demand

Supply

Demand

Supply

Demand

Supply

Week 1

Week 2

Week 3

Week 4

Approach 1

Approach 2

Approach 3

Approach 4

Approach 5

Another strategic decision on feeder lines is the ratio between owned and chartered

ships in the fleet. Feeder lines usually operate a small fixed number of owned ships and

balance their requirements with chartered ships. This could decrease their capital costs

and make their network more flexible to changes in trade. However, if there is a stable

or increasing demand trend in the market, operating with a high number of charter ships

could be couple of times more costly than operating with owned ships. Therefore, it is

crucial for feeder lines to define the minimum number of owned feeder ships for long

term efficiency.

In this study, we compared six different approaches in order to determine owned

ship numbers at the start of planning horizon (Table 7.21). The first approach starts with

no owned ship and uses only charter in ships during the planning horizon. The second

approach assigns ships according to maximum slot demanded season of the planning

horizon in the no owned ship approach. This approach guarantees that all demand and

supply on the market will be satisfied by the shipping lines. However, it could result in

overcapacity in the total transportation slots. During the planning periods, when the

ships become idle, they could be chartered out. The third approach defines owned ship

numbers according to the maximum necessary ship numbers from each type of ship

from all seasons which is calculated in the no owned ship approach. This approach also

guarantees satisfaction of the customer demand, but could decrease utilization of the

owned on hand ships. The forth approach assigns ships according to minimum slot ca-

pacity demanded season of the planning horizon in the no owned ship approach. This

approach does not guarantee the satisfaction of the demand of the ports, but increases

the utilization of the ships and charter in ship numbers. The fifth approach defines

owned ship numbers according to the minimum necessary ship numbers from each type

7. Numerical Investigation 129

of ship from all seasons which is calculated in the no owned ship approach. This ap-

proach also does not guarantee satisfaction of the customer demand, but could decrease

total owned ship costs and increase charter in ship numbers. The sixth approach defines

on hand ships according to average necessary ship numbers from each type of ship dur-

ing the planning horizon which is defined in the no owned ship number approach.

Table 7.22 shows an example of owned ship number determination approaches for

four periods and three ship types.

Table 7.21: Owned ship number determination approaches (Scenario C)

Approach

No owned ship

According to maximum slot demanded season of the network in all seasons

According to maximum necessary ship numbers from each type of all seasons

According to minimum slot demanded season of the network in all seasons

According to minimum necessary ship numbers from each type of all seasons

According to average necessary ship numbers from each type of all seasons

Table 7.22: Example of owned ship number determination approaches

Ship 1 (4300)

Ship 2 (2600)

Ship 3 (1200)

Slot capacity

Approach 1

Charter

out

TEU

On hand

Period 1

22400

Period 2

19000

Period 3

30700

Period 4

15000

Approach 2

On hand

30700

Period 1

22400

Period 2

19000

Period 3

30700

Period 4

15000

Approach 3

On hand

31900

Period 1

22400

Period 2

19000

Period 3

30700

Period 4

15000

Approach 4

On hand

15000

Period 1

22400

Period 2

19000

Period 3

30700

Period 4

15000

Approach 5

On hand

13800

Period 1

22400

Period 2

19000

Period 3

30700

Period 4

15000

Approach 6

On hand

22600

Period 1

22400

Period 2

19000

Period 3

30700

Period 4

15000

7. Numerical Investigation 130

In addition to the previous experiments, a comprehensive numerical investigation is

presented to evaluate how the general change in ship prices (owning, charter in, charter

out) impacts on the design of the service network configuration.

Table 7.23: The change in the ship prices (Scenario E)

Approach

50% decrease in ship prices

25% decrease in ship prices

No change in ship prices

25% increase in ship prices

50% increase in ship prices

100% increase in ship prices

In order to evaluate the performance of the week allocation approaches, owned ship

number approaches, and demand allocation approaches with service environment pa-

rameters an experimental design framework is conducted. The conducted experimental

design consists of four test families by using four design and environment scenarios

(Table 7.24).

Table 7.24: Designed experimental tests

Test

Scenario A

Scenario B

Scenario C

Scenario D

1-2-3-4-5-6

1-2-3-4-5

1-2-3-4-5-6

7.4.6 Numerical results

The designed experimental tests in the previous section are solved by using the de-

veloped ANS algorithm for a 52-week sailing period with the forecasted demands in

Section 7.4.3 Please see Total throughput of related container terminals, the operation amount of interest-

ed feeder shipping line could be calculated by using market share ratios in Table A.10.

Table A.13 in the Appendix for the detailed results of the experiments. Figure 7.16

shows how the weekly network costs of a different number of periodical seasons change

during the sailing period. According to the results, as long as the periodical change

number increases, the flexibility of the service network increases in order to adapt itself

to seasonal demand fluctuations. The total cost of the network therefore increases when

the seasonal period number is decreased in a sailing period. When monthly change in

the service network is allowed, the total network cost for a sailing season is around 309

million dollars. Total network cost is around 316 million dollars for bi-monthly change,

320 million dollars for quarterly change, 323 million dollars for trimester change, 324

7. Numerical Investigation 131

million dollars for semi-annual change, and 340 million dollars for annual seasonal

change to update demand figures and service network designs. Figure 7.17 shows how

the slot utilization of the ships changes for different numbers of periodical seasons dur-

ing weeks in the sailing season. As long as the number of change periods is increased,

the utilization ratio of the ship slots increases as well. Both cost and utilization figures

show the importance of flexible and demand dependent service network designs. How-

ever, despite the advantage of changing the design of service networks more frequently,

it is not practical to change service networks every two months or sooner. Therefore

under assumptions of this scenario changing a service network quarterly is the most

effective approach for shipping lines under demand fluctuations.

Figure 7.16: Effect of seasonal change number on the weekly total cost of the network

(Test 1)

Figure 7.17: Effect of seasonal change number on the weekly capacity utilization of the

network (Test 1)

7. Numerical Investigation 132

Figure 7.18 shows the result of the second test family which evaluates the dual im-

pact of different numbers of periodical seasons and different seasonal demand assign-

ment approaches by considering other scenarios fixed. Figure 7.19 shows how the slot

utilization of the ships changes for different demand assignments during weeks in the

sailing season by considering quarterly service network change. In these figures, the

network cost of rule 1 and rule 2, which use maximum based demand approaches, high-

ly increases as the number of network changes is decreased because of the usage of high

demanded week’s rates. Since these approaches are allocating more slots to terminals,

they require more ships during the planning periods. By designing according to maxi-

mum demanded week’s values, the network design ensures that there will be no overca-

pacity during the weeks. But this will come with respectively low utilization rates. On

the other hand in minimum based demand approaches (3 and 4), the total network cost

decreases as the number of network change is decreased because of the usage of low

demanded week’s values. Since these approaches are allocating fewer slots to terminals,

they require fewer ships during the planning periods. By designing according to mini-

mum demanded week’s values, the network design aims to increase utilization rates

during the weeks. However this could result in overcapacity in rush weeks. Although

the third approach, which assigns the minimum demanded week of each terminal in the

season as seasonal demand, has the lowest network costs of all seasonal change ap-

proaches, its utilization rate is always more than 100% which could result in lost sales

during the periods. The forth approach which assigns minimum total demanded week’s

values as seasonal demand also has correspondingly low total network costs. But its

utilization rate is generally more than 100% which could also cause lost sales during the

planning horizon. The cost of the fifth demand assignment approach is almost the same

in all seasonal change approaches. However robustness of it disappears when the

change number is decreased. It starts to become more over capacity. The approach also

generally has more than 90% capacity utilization which increases the risk of overcapaci-

ty. In practice, the expected capacity utilization is around 90% in order to handle other

unexpected demand fluctuations and irregular container types. Therefore it could be

suggested that the second demand assignment approach, which assigns according to

maximum total demanded week of the season, has more robust conditions to handle

demand fluctuations within the season. Thus, demand change scenario 2 and seasonal

change scenario 3 is the most effective and robust combination under these experi-

mental conditions.

7. Numerical Investigation 133

Figure 7.18: Impact of seasonal change number and demand assignment (Test 2)

Figure 7.19: Impact of demand assignment on the weekly capacity utilization of the

network (Test 2)

Figure 7.20 shows the result of the third test family which evaluates the dual impact

of different numbers of periodical seasons and different starting owned ship number

determination approaches by considering other scenarios fixed. The figure clearly

shows the importance of owned ship number on decreasing total cost of the network. In

all service network change approaches, the no owned ship approach has an almost 20%

cost disadvantage compare to other approaches which use different owned ship strate-

gies. As the service network change number decreases, the impact of owned ship num-

ber also decreases in these configurations. Within owned ship number approaches, max-

imum ship and slot based approaches (2 and 3) have the lowest total network costs. On

the other hand, the network cost is slightly diverse between maximum and minimum

based approaches. Therefore, the minimum demanded slot based owned ship approach

could be used in order to decrease capital investment and become flexible to fluctua-

tions in the seasonal demands. Please note that the approaches in this test assume that

there is a broad market to charter in and out ships with no trouble.

7. Numerical Investigation 134

Figure 7.20: Impact of seasonal change number and owned ship number (Test 3)

Figure 7.21 shows the result of the fourth test family which analyzes the dual effect

of different starting owned ship number determination approaches and the change in the

ship prices by considering other scenarios fixed. The rates in the figures show that as

long as ship prices increase the impact on owned ship number increases in the total net-

work cost. With low ship prices the total network cost starts to be less ship cost oriented

and the number of owned ships loses its importance. On the other hand, the total net-

work cost is more ship price oriented under high ship prices.

Figure 7.21: Impact of the owned ship number and ship price (Test 4)

7.4.7 Concluding remarks

The decisions in tactical planning operations of liner shipping, such as fleet size and

mix, fleet deployment, ship routing and scheduling are made according to container

throughput estimations. However, these throughputs are highly affected by unstable

financial, political and seasonal conditions. Therefore in this study, an adaptive neigh-

borhood search approach is used to determine the feeder ship fleet size and mix, fleet

deployments, service routes and voyage schedules to minimize operational costs for

dynamic sailing seasons. A Monte Carlo simulation and an artificial neural networks

based forecasting framework are also used to estimate unstable throughput demands of

regional ports. In our case study investigation, we assume the feeder network design

problem of a Turkish short-sea shipping company in view of the opening of the new

Candarli port near Izmir. The cost performance of alternate feeder network configura-

tions serving the Black Sea region is compared under unstable demand environments.

7. Numerical Investigation 135

The optimal feeder networks are calculated according to the forecasted throughputs of

the region terminals for each week during a 52-week sailing season. The results show

that the total cost of the region service network is affected around 34.21%, the total slot

capacity is affected around 47.64%, and utilization of the slot capacity, the necessary

ship type and mix are highly affected by these unstable conditions. These figures show

the importance of dynamic and flexible feeder service network designs in liner shipping

in making effective and efficient plans. This study could be extended by developing

liner shipping service network designs under unstable freight rate and oil price envi-

ronments.

8. Summary 136

8. Summary

The introduction of mega containerships on the main international sea routes between

major seaports made it necessary to temporarily store containers in a specific region and

to distribute them on short-sea routes. Therefore in addition to the location of hub ports,

regional feeder containership service is a critical issue in designing global networks of

shipping lines. In conceptual terms, the feeder containership service collects and drops

containers in a specific region with small and medium sized containerships and feeds

mega containerships so as to avoid their calling at too many ports. It was the container-

ship feeder line that made the entire container service economically rational, efficient,

more profitable, and consequently cheaper and timely for the end users. Hence devel-

opments of effective service network designs are essential in order to better help deci-

sion makers of liner shipping providers in different environments. In a sailing season, an

effective service network includes joint solution of tactical planning decisions, such as

fleet size and mix, fleet deployment, ship routing and scheduling. The container feeder

network design depends on the characteristics of feeder ships, the feeder ship ports, the

operating and chartering costs of the ships and bunker costs, as well as container de-

mand throughputs of the ports.

The decisions in tactical planning operations of liner shipping, such as fleet size and

mix, fleet deployment, ship routing and scheduling are made according to container

throughput estimations. Container demand throughputs of ports are therefore one of the

main design parameters of service networks. Nevertheless, they have been directly af-

fected by unexpected local and global demand fluctuations as well as seasonal condi-

tions. Therefore, forecasting container throughputs of ports is playing a critical role at

all the levels of planning decisions of liner shipping lines. Since liner shipping involves

considerable capital investment and huge daily operating costs, the appropriate through-

put demand estimation of a whole sailing season will state the development of service

network design. In order to cope with the dynamic nature of shipping markets, it is also

important to design more agile and flexible feeder service networks by allowing season-

al service network changes within a planning horizon.

8. Summary 137

Changes to the service network may include introducing new routes, and schedules

as well as fleet deployments which could contain chartering in new ships or chartering

out unnecessary ships. This thesis also employs various service scenarios in order to

better help decision makers of liner shipping providers. These scenarios contain differ-

ent periodical approaches, different demand allocations, different numbers of owned

ships at the start of sailing season, and different ship prices to provide a very high de-

gree of ﬂexibility to planning decisions under unstable demand environments.

In Chapter 2 the reader is introduced to container shipping and especially to con-

tainerization history, terminology and standards, orientation of liner shipping, leading

shipping lines and freight rates, and common service network designs.

Detailed information about the background of liner shipping feeder services, ad-

vantages and disadvantages of feeder services, the role of feeder services in modern

global service networks, design parameters of feeder service networks, characteristics of

feeder lines, the reasons for demand fluctuations in feeder services, and general problem

types in different planning levels are presented in Chapter 3.

In Chapter 4 the relevant literature is summarized in order to underline potential

gaps in liner shipping network designs, feeder services, vehicle routing problems, con-

tainer throughput estimation, and liner shipping under unstable demand environments.

The feeder service network design (FND) problem is mathematically modeled in

four parts in Chapter 5. While Section 5.1 handles the problem as the vehicle routing

problem with simultaneous pick-up and delivery with time limit (VRPSPDTL), Section

5.2 handles the problem as feeder containership routing problem (FCRP) which is the

basic version of FND without including sailing season, ship economies, and ship mix

and deployment. Section 5.3 handles the basic FND problem for a stable sailing horizon

for reducing the total transportation costs and Section 5.4 approaches the problem more

realistically by considering varying forecasted throughput demands for a dynamic sail-

ing season and vessel charter operations.

Two frameworks for the feeder service network design problems are provided in

Chapter 6. The first framework proposes an Adaptive Neighborhood Search (ANS)

combined with the classic savings heuristic as an initial solution construction algorithm,

variable neighborhood search in order to improve the initial solution, and a perturbation

8. Summary 138

mechanism to escape from local optima. The second framework provides a Monte Carlo

simulation and an artificial neural networks based forecasting frame in order to analyze

the impact of seasonal demand fluctuation on the liner shipping feeder service.

Numerical results of FND related problems are presented in four sections in Chapter

7. In Section 7.1, well known benchmark problem instances of the vehicle routing prob-

lem with simultaneous pick-up and delivery with time limit (VRPSPDTL) are solved

with the developed ANS approach. From the numerical results it can be concluded that

the proposed ANS algorithm generates efficient and robust solutions compared to exist-

ing solution methods for the VRPSPDTL. The ANS approach could obtain new best

solutions or reach the best known solution for 19 out of the 28 benchmark instances

with and without service time. This section proved the robustness and effectiveness of

the developed ANS approach for vehicle routing problems. The proposed ANS algo-

rithm can be adapted to consider heterogeneous fleet conditions and the dynamic envi-

ronment of the VRPSPDTL.

In Section 7.2, FCRP with a case study from the Aegean Islands are solved with the

developed ANS approach. The solutions existing in the literature improved around 3%

by using the developed approach. Moreover, a range of scenarios and parameter values

are used in order to test the robustness of the approach through sensitivity analysis.

From the numerical results it can be concluded that the proposed ANS algorithm gener-

ates efficient and robust solutions for the FCRP. This section shows effective imple-

mentation of ANS to containership routing problems. The proposed ANS algorithm can

be adapted in order to consider heterogeneous fleet conditions, multi depot characteris-

tics, and dynamic environments of the FCRP.

In Section 7.3, the FND problem which aims to simultaneously determine the fleet

size and mix, fleet deployment, ship routing and ship scheduling by minimizing total

network costs in a sailing season are solved by using the developed ANS approach. In

this section, we focus on the potential hub role of a new port (Candarli) in the East

Mediterranean and Black Sea region for a short-sea shipping company. In the numerical

investigation the cost performance of three strategic options for hub port configurations

has been compared. From the numerical results it can be concluded that Candarli as a

new hub port offers significant cost savings compared to Port Said which is currently

used as a hub port by the considered company. However, these cost savings would be

compensated with additional transhipment cost for the Port Said - Candarli services

8. Summary 139

which are needed to connect Candarli to the global trunk shipping lines. Therefore, ad-

ditional factors like service quality and handling efficiency at the hub ports as well as

waiting time in the queue of the hub ports play an important role in the development of

the company's feeder network configuration. Certainly, the new Candarli port has great

market potential as long as port authorities keep container handling costs and service

quality at a favourable level.

Finally, Section 7.4 considered the service network design problem of liner shipping

networks under unstable demand environments from the perspective of a feeder ship-

ping company commencing its services from a newly constructed port. The design of

service networks is revised at the beginning of every period in response to changes in

demand patterns for a season estimated with a simulation based forecasting framework.

The proposed model implementation has been tested for a liner shipping feeder service

in the Black Sea region. The optimal feeder networks are designed by using the fore-

casted throughputs of the region terminals for each week during a 52-week sailing sea-

son with the help of the developed ANS approach. The results show that the total cost of

the region service and the total slot capacity is highly affected by unstable demand con-

ditions. The related section also handles various service scenarios such as different peri-

odical approaches, different demand allocations, different number of owned ships at the

start of sailing season, and different ship owning, chartering, and oil prices in order to

better help decision makers of liner shipping providers. These figures show the im-

portance of dynamic and flexible feeder service network designs in liner shipping to

make effective and efficient plans. This study could be extended by developing liner

shipping service network designs under unstable freight rates and oil prices and envi-

ronmental routing conditions.

Appendices 140

A. Appendices

Table A.1: The top 100 liner shipping operators in 2013

Shipping Line

Slot

World

Total

Average

Rank

Operator

TEU

Ships

TEU

APM-Maersk

2,581,417

15.68%

593

4353.15

Mediterranean Shg Co

2,295,334

13.94%

473

4852.71

CMA CGM Group

1,421,398

8.63%

419

3392.36

COSCO Container L.

730,598

4.44%

158

4624.04

Evergreen Line

721,684

4.38%

184

3922.20

Hapag-Lloyd

652,750

3.96%

141

4629.43

APL

607,326

3.69%

128

4744.73

Hanjin Shipping

602,536

3.66%

114

5285.40

CSCL

569,186

3.46%

139

4094.86

MOL

499,893

3.04%

108

4628.64

OOCL

461,259

2.80%

4706.72

NYK Line

414,299

2.52%

4361.04

Hamburg Süd Group

413,498

2.51%

102

4053.90

K Line

352,106

2.14%

4959.24

Yang Ming Marine

Transport Corp.

350,646

2.13%

4328.96

Hyundai M.M.

346,097

2.10%

5967.19

Zim

320,018

1.94%

3902.66

PIL (Pacific Int. Line)

300,133

1.82%

146

2055.71

UASC

279,460

1.70%

5703.27

CSAV Group

255,536

1.55%

4646.11

Wan Hai Lines

162,501

0.99%

2288.75

HDS Lines

86,320

0.52%

4110.48

X-Press Feeders Group

76,466

0.46%

1274.43

NileDutch

67,482

0.41%

2176.84

TS Lines

67,342

0.41%

1980.65

SITC

65,892

0.40%

1029.56

KMTC

54,936

0.33%

1408.62

RCL (Regional Con-

tainer L.)

50,209

0.30%

1394.69

CCNI

43,846

0.27%

2579.18

STX Pan Ocean (Con-

tainer)

42,722

0.26%

1941.91

Grimaldi (Napoli)

39,944

0.24%

1024.21

UniFeeder

39,821

0.24%

971.24

Sinotrans

38,673

0.23%

1208.53

Seaboard Marine

36,690

0.22%

1048.29

Arkas Line / EMES

36,428

0.22%

1103.88

Matson

33,231

0.20%

1444.83

Simatech

33,122

0.20%

2208.13

Meratus

30,229

0.18%

570.36

Samudera

29,805

0.18%

805.54

Salam Pasific

28,327

0.17%

590.15

OEL / Shreyas

(Transworld Group)

27,767

0.17%

1262.14

Schöller Group

27,447

0.17%

1715.44

Horizon Lines

26,264

0.16%

2020.31

Heung-A Shipping

26,166

0.16%

1006.38

Linea Messina

25,154

0.15%

1934.92

S.C. India

24,467

0.15%

3495.29

Tanto Intim Line

24,012

0.15%

600.30

Swire Shipping

23,918

0.15%

1087.18

Quanzhou An Sheng

Shg Co

23,078

0.14%

824.21

Appendices 141

Shipping Line

Slot

World

Total

Average

Rank

Operator

TEU

Ships

TEU

Sinokor

22,599

0.14%

903.96

MACS

22,396

0.14%

1599.71

FESCO

21,147

0.13%

1057.35

Nam Sung

21,112

0.13%

879.67

Mariana Express Lines

20,020

0.12%

1430.00

Crowley Liner Services

18,718

0.11%

935.90

DAL

18,138

0.11%

2267.25

Log-In Logistica

16,402

0.10%

2050.25

Turkon Line

15,578

0.09%

1730.89

King Ocean

14,964

0.09%

880.24

Westwood

14,699

0.09%

2099.86

Grand China Logistics

14,693

0.09%

1469.30

Temas Line

14,660

0.09%

542.96

Hainan P O Shipping

13,937

0.08%

1991.00

Peel Ports (BG Freight)

13,222

0.08%

881.47

Great White Fleet

13,082

0.08%

568.78

United Feeder Services

12,887

0.08%

991.31

Emirates Shipping Line

12,630

0.08%

2526.00

Dole Ocean Liner

12,319

0.07%

684.39

Borchard Lines

12,190

0.07%

870.71

Marfret

12,089

0.07%

1343.22

Caribbean Feeder Ser-

vices

10,382

0.06%

741.57

Interasia Line

10,230

0.06%

1461.43

Delphis NV / Team

Lines

10,030

0.06%

911.82

Containerships OY

9,480

0.06%

790.00

Independent Container

Line

9,250

0.06%

2312.50

Goto Shipping

9,033

0.05%

1505.50

Vinalines

8,485

0.05%

606.07

SASCO (Sakhalin

Shipping Co)

8,395

0.05%

399.76

Shanghai Jin Jiang

7,794

0.05%

866.00

Chun Kyung (CK)

Line)

7,139

0.04%

509.93

Melfi C.L.

6,878

0.04%

1146.33

Lin Line

6,672

0.04%

2224.00

Eimskip

6,468

0.04%

646.80

SeaFreight

6,376

0.04%

1062.67

Tarros

6,343

0.04%

1268.60

Shin Yang Shipping

Sdn Bhd

6,219

0.04%

365.82

Tropical Shg

6,178

0.04%

475.23

Kambara Kisen

6,136

0.04%

767.00

Caraka Tirta Perkasa

6,103

0.04%

678.11

Samskip

6,100

0.04%

610.00

Qatar Navigation

(Milaha)

6,095

0.04%

761.88

HubLine Bhd

5,956

0.04%

595.60

Shanghai Hai Hua

(Hasco)

5,919

0.04%

739.88

Maestra Navegaçao

5,674

0.03%

1418.50

OPDR

5,636

0.03%

626.22

Boluda Lines

5,427

0.03%

904.50

Merchant Shipping

5,387

0.03%

2693.50

Valfajre Eight Shg Co

5,299

0.03%

662.38

Perkapalan DZ PDZ

Lines)

4,793

0.03%

684.71

100

TransAtlantic AB

4,770

0.03%

477.00

Total

Top 3

6,298,149

38.25%

1485

4241.1778

Total

Top 5

7,750,431

47.07%

1827

4242.1626

Total

Top 10

10,682,122

64.88%

2457

4347.628

Total

Top 20

14,175,174

86.10%

3294

4303.3315

Total

Top 50

15,490,032

94.08%

4257

3638.7202

Total

Top 100

16,013,562

97.26%

4810

3329.2229

Total

World

16,464,087

100.00%

4953

3324.0636

Source: Alphaliner (2013b)

Appendices 142

Table A.2: Computer / Software benchmarking results

Algorithm

Computer

Softwarec

Scorea

Ratiob

CIH

VAX 4000-500 71.4 MHz

Fortran 77

10*

0.01

ALT

Intel Pentium II 333 MHz

Fortran 77

25*

0.02

LNS

Intel Pentium IV 1.5 GHz 256 MB

RAM

C++

173

0.15

AMD Athlon XP 2.0 GHz 256 MB

RAM

Pascal Deplhi 5.0

360

0.32

RTS

Sun Fire V440 1062 MHz

Fortran

100*

0.09

ILS

Intel Core 2 Duo 2.13 GHz 1024

MB RAM

C++ 6.0

850

0.76

ACS

Intel Xeon 2.4 GHz

350*

0.31

ANS

Intel Core 2 Duo T5720 2.0 GHz 2

GB RAM

Matlab R2009a &C#

2010

1117

1.00

a Computer performance scores are according to Passmark Performance Test 7.0 software; b: Best solution times algorithms are

scaled according to make a fair comparison. * Approximated values.

Table A.3: Best solution route sequences for CMT10Y* with service time

Route

Number

Route

Sequence

Route

Distance

Route

Duration

Starting

Load

Returning

Load

0-9-135-35-136-65-71-161-103-51-0

104.6

194.63

18.94

126.06

0-180-198-197-56-186-39-187-139-4-155-110-

149-0

75.15

195.15

129.30

71.70

0-179-130-165-55-25-170-67-23-75-72-0

98.82

198.82

124.26

42.74

0-33-81-120-164-34-78-169-29-121-68-184-0

86.81

196.81

30.05

169.95

0-59-193-91-191-44-140-38-14-119-192-100-0

89.71

199.71

23.39

136.61

0-124-168-47-36-143-49-64-11-0

112.6

192.63

100.40

16.60

0-132-30-160-128-66-188-20-122--1-176-111-

86.63

196.63

48.39

137.61

0-13-117-151-92-37-98-85-93-99-104-96-6-

183-112-0

51.67

191.67

32.30

182.70

0-31-108-90-126-63-181-32-131-70-101-69-0

89.88

199.88

70.55

76.45

0-94-95-97-87-172-42-142-43-15-57-144-137-

79.74

199.74

56.26

105.74

0-50-102-157-185-79-129-158-3-77-116-196-

76-28-0

58.52

188.52

24.85

186.15

0-27-167-127-190-88-148-182--7-194-106-

153-52-146-0

54.28

184.28

86.95

113.05

0-147-5-173-61-16-141-86-113-17-84-118-0

85.38

195.38

120.91

103.09

0-162-189-10-159-62-175-107-19-123-48-82-0

84.18

194.18

134.68

79.32

0-105-40-21-73-171-74-133-22-41-145-115-

178-2-0

67.50

197.50

114.23

73.77

0-156-152-58-53-0

20.35

60.35

15.05

61.95

0-18-114-8-174-46-45-125-199-83-60-166-89-

78.18

198.18

111.32

55.68

0-26-195-54-134-24-163-80-150-177-109-12-

138-154-0

66.86

196.86

73.09

131.91

Total

1390.92

*Vehicle capacity (Q): 200; Time Limit (R): 200

Appendices 143

Table A.4: Best solution route sequences for CMT13X* without service time

Route

Number

Route

Sequence

Route

Distance

Route

Duration

Starting

Load

Returning

Load

0-95-109-37-38-39-42-41-44-47-46-49-50-51-

48-43-40-68-76-77-79-80-78-72-71-70-69-67-

103-104-107-106-105-120-0

266.37

190.90

197.09

0-102-101-99-100-116-98-110-115-97-94-96-

93-92-89-84-113-83-117-112-85-87-86-111-

82-119-0

89.17

120.47

159.52

0-52-57-54-53-55-58-56-60-63-66-64-62-61-

65-59-45-29-32-28-35-36-34-31-30-33-27-24-

22-25-19-16-17-20-23-26-21-0

322.80

199.17

185.82

0-91-90-114-18-118-108-8-12-13-14-15-11-

10-9-7-6-5-4-3-1-2-81-88-0

138.53

71.56

195.43

Total

816.87

*Vehicle capacity (Q): 200; Time limit (R): 720

Table A.5: Demand, supply, service time of feeder ports and distances between ports*

Name

Demand

Supply

Servive

time

Pireaus

Rethimno

Iraklion

Kithira

Milos

Kea

Thira

Paros

Naxos

Syros

Mykonos

Tinos

Andros

Chios

Ikaria

Samos

Astypalea

karpathos

Rhodes

Kos

Kalimnos

Skyros

Mytilini

Limnos

Amorgos

Sifnos

158

174

124

146

103

153

143

190

193

287

298

238

219

144

215

221

148

1,8

158

108

100

165

130

136

152

157

162

182

237

194

226

139

177

233

200

186

251

300

330

133

118

6,1

174

142

109

174

126

129

152

159

184

227

179

206

115

141

199

171

161

258

289

328

117

122

1,3

124

108

142

122

138

135

154

138

160

153

151

233

211

251

194

251

302

251

234

216

294

306

169

107

3,1

100

109

153

130

164

113

185

221

166

148

152

216

235

1,3

165

174

122

106

119

133

153

152

229

254

178

160

165

166

105

2,4

146

138

106

119

144

131

128

166

116

100

178

225

247

2,4

130

126

135

109

115

164

183

124

106

131

172

202

1,7

103

136

129

154

101

104

157

171

111

135

164

201

1,8

152

138

120

119

192

203

142

123

104

152

177

2,5

157

152

160

180

182

120

102

111

142

178

1,4

162

159

153

105

100

185

194

132

113

100

138

169

1,2

182

184

151

119

103

129

152

223

243

164

146

135

145

101

2,2

153

237

227

233

153

119

144

109

101

133

216

207

126

114

107

124

113

130

1,2

143

194

179

211

130

133

103

161

162

133

135

177

103

3,2

190

226

206

251

164

153

131

115

104

120

105

129

163

158

159

146

185

144

1,4

193

139

115

194

113

152

119

100

152

133

109

195

223

260

103

1,4

287

177

141

251

185

229

128

164

157

192

180

185

223

216

161

163

100

106

286

310

350

124

183

3,9

298

233

199

302

221

254

166

183

171

203

182

194

243

207

162

158

109

288

317

360

139

212

2,0

238

200

171

251

166

178

116

124

111

142

120

132

164

126

100

217

225

284

152

1,4

219

186

161

234

148

160

100

106

123

102

113

146

114

106

201

211

271

133

1,1

144

251

258

216

152

178

131

135

104

111

100

107

133

159

195

286

288

217

201

115

164

134

3,7

215

300

289

294

216

165

225

172

164

152

142

138

135

146

223

310

317

225

211

115

102

197

199

2,1

221

330

328

306

235

166

247

202

201

177

178

169

145

124

177

185

260

350

360

284

271

102

230

213

1,2

148

133

117

169

105

101

113

124

139

164

197

230

1,2

118

122

107

130

103

144

103

183

212

152

133

134

199

213

*Demand & supply (small container), service (hours), distances (nautical miles) (Karlaftis et al. 2009)

Appendices 144

Table A.6: Route details of best solution for soft time deadline

Route

number

Route

sequence

Starting

load

Return-

ing load

Last

approaching

time

Total

route

delay

Route

duration

Fitness

value

0-11-18-17-16-8-0

100

52.27

12.27

62.55

63.16

0-5-9-10-7-25-4-0

25.28

36.47

0-12-21-23-22-13-0

40.35

0.35

55.30

55.32

0-6-24-20-19-15-14-0

39.70

52.82

0-3-1-2-0

25.60

46.20

Total

464

235

12.62

253.34

253.96

Table A.7: Route details of best solution for hard time deadline

Route

number

Route

sequence

Starting

load

Return-

ing load

Last

Approaching

Time

Total

Route

Delay

Route

duration

Fitness

value

0-8-16-17-18-0

33.92

62.65

0-5-12-9-11-10-7-25-

4-0

29.50

40.73

0-3-2-1-0

39.70

52.82

0-6-24-20-19-15-14-0

25.60

46.20

0-21-23-22-13-0

38.65

53.60

Total

464

235

256.00

Table A.8: Demand parameters for related container terminals*

No.

Terminal

Port1

Region

Supply2

Demand3

Handling2

Port Said

Black Sea

Candarli (Izmir)

Black Sea

Burgas

Black Sea

16.7

Varna

Black Sea

Constanta 1

Black Sea

Constanta 2

Black Sea

100

105

16.7

Illiychevsk

Black Sea

33.3

Odessa

Black Sea

120

160

16.7

Novorossiysk 1

Black Sea

166

110

Novorossiysk 2

Black Sea

108

12.5

Poti

Black Sea

112

11.1

Batumi

Black Sea

Trabzon

Black Sea

11.1

Haydarpasa (Istanbul)

Sea of Marmara

33.3

Ambarli 1 (Istanbul)

Sea of Marmara

179

234

33.3

Ambarli 2 (Istanbul)

Sea of Marmara

120

166

33.3

Ambarli 3 (Istanbul)

Sea of Marmara

105

Gebze 1 (Izmit)

Sea of Marmara

33.3

Gebze 2 (Izmit)

Sea of Marmara

Gemlik 1 (Bursa)

Sea of Marmara

33.3

Gemlik 2 (Bursa)

Sea of Marmara

Gemlik 3 (Bursa)

Sea of Marmara

33.3

Aliaga 1 (Izmir)

Aegean Sea

Aliaga 2 (Izmir)

Aegean Sea

33.3

İzmir

Aegean Sea

165

Thessaloniki

Aegean Sea

33.3

Piraeus 1

Aegean Sea

111

33.3

Piraeus 2

Aegean Sea

102

189

Antalya

East Med. sea

Mersin

East Med. sea

187

Limassol

East Med. sea

Lattakia

East Med. sea

33.3

Beirut

East Med. sea

Appendices 145

No.

Terminal

Port1

Region

Supply2

Demand3

Handling2

Haifa

East Med. sea

107

142

Ashdod

East Med. sea

123

125

Alexandria 1

East Med. sea

130

33.3

Alexandria 2

East Med. sea

33.3

Damietta

East Med. sea

132

* Distances between ports calculated by use of the Netpas Distance software; 1 Port code of terminal as shown in Figure 5; 2 Termi-

nal’s daily container supply in TEU; 3 Terminal’s daily container demand in TEU;4 Terminal container handling efficiency in TEU

per hour.

Table A.9: Best solution for the feeder network design of the Candarli port

Rout

e no.

Port

sequence

Total

costs

Total

de-

mand

Total

supply

Ship type

(TEU)

Required

ships

Service

no.

Voyage

duration

On sea

dura-

tion

On port

dura-

tion

0-25-26-24-0

2.27E+0

1330

2492

2600

253.29

34.62

218.67

0-27-33-35-34-0

2.86E+0

1463

2499

2600

299.62

80.20

219.42

0-14-0

8.38E+0

840

1162

1200

130.69

27.59

103.10

0-21-13-17-0

2.23E+0

1736

2590

2600

254.53

29.70

224.83

0-31-30-28-0

2.64E+0

1302

2429

2600

295.67

72.66

223.01

0-23-0

7.32E+0

602

1155

1200

114.86

6.44

108.42

0-22-19-18-0

1.08E+0

763

1085

1200

137.3

30.29

107.01

0-12-6-4-2-1-0

3.59E+0

1988

2569

2600

417.82

69.20

348.62

0-7-8-5-3-0

3.65E+0

2597

2429

2600

425.32

93.27

332.05

0-11-10-9-20-0

2.02E+0

455

1197

1200

271.09

103.85

167.24

0-15-16-0

9.83E+0

721

1176

1200

144.55

33.16

111.39

0-29-32-36-0

2.63E+0

1260

2464

2600

290.61

72.71

217.90

Total

2.55E+0

15057

23247

24200

624

3035.34

0.16

0.58

Rout

e no.

Feeder

port dura-

tion

Hub port

duration

Lay-up

duration

Idle

dura-

tion

Total

costs

Opera-

tion cost

ratio

Charter

cost ratio

Adminis-

trative

cost ratio

On sea

cost ratio

On port

cost

ratio

Port

cost

ratio

140.43

78.24

24.00

58.71

2.27E+0

18.28%

24.27%

10.18%

14.46%

6.28%

26.53%

138.38

81.04

24.00

12.38

2.86E+0

14.54%

19.31%

8.10%

26.65%

5.01%

26.39%

61.56

41.54

16.80

20.51

8.38E+0

20.17%

25.48%

6.00%

16.74%

4.30%

27.31%

136.51

88.32

24.00

57.47

2.23E+0

18.63%

24.74%

10.38%

12.64%

6.58%

27.04%

146.59

76.42

24.00

16.33

2.64E+0

15.75%

20.92%

8.78%

26.16%

5.52%

22.87%

71.78

36.64

16.80

36.34

7.32E+0

23.08%

29.16%

6.86%

4.47%

5.17%

31.25%

68.55

38.46

16.80

13.90

1.08E+0

15.62%

19.74%

4.64%

14.24%

3.46%

42.30%

255.68

92.94

24.00

62.18

3.59E+0

17.37%

23.07%

9.68%

18.31%

6.34%

25.22%

229.73

102.32

24.00

54.68

3.65E+0

17.05%

22.65%

9.50%

24.23%

5.93%

20.63%

132.70

34.54

16.80

48.11

2.02E+0

16.70%

21.10%

4.96%

26.09%

2.89%

28.26%

71.95

39.44

16.80

6.65

9.83E+0

17.18%

21.71%

5.11%

17.14%

3.96%

34.90%

141.62

76.28

24.00

21.39

2.63E+0

15.77%

20.94%

8.79%

26.21%

5.40%

22.90%

Total

0.39

0.19

0.06

0.10

2.55E+0

16.99%

22.30%

8.43%

20.83%

5.38%

26.07%

Table A.10: Figures of contracted container terminals*

Terminal

Country

M.share**

2005***

2006

2007

2008

2009

2010

2011

Burgas

Bulgaria

31.00%

25000

26400

30600

45900

23800

23500

25000

Varna

Bulgaria

21.00%

84000

94000

99700

155300

112600

118700

122844

Constanta 1

Romania

19.00%

476600

737100

1111400

1080900

294300

256500

350000

Constanta 2

Romania

24.00%

300000

Illiychevsk

Ukraine

20.00%

291100

312100

532800

670600

256800

301500

280000

Odessa

Ukraine

22.00%

288400

395600

523500

572100

255500

354500

453700

Novorossiysk 1

Russia

29.00%

60000

90100

182000

84000

188652

335847

Novorossiysk 2

Russia

29.00%

99100

141400

124500

111000

124626

200153

Poti

Georgia

20.00%

105900

126900

184800

209600

172800

209800

254022

Batumi

Georgia

20.00%

44200

8800

16300

45439

Trabzon

Turkey

35.00%

300

5400

22300

22100

21100

34072

40251

Haydarpasa

Turkey

15.00%

340600

400100

369600

356300

191400

176500

206082

Ambarli 1

Turkey

10.00%

790300

962900

1296800

1541200

1263600

1663600

1548485

Ambarli 2

Turkey

15.00%

439000

531000

666000

649000

476000

621000

844000

Appendices 146

Terminal

Country

M.share**

2005***

2006

2007

2008

2009

2010

2011

Ambarli 3

Turkey

15.00%

161500

198500

276300

359700

200200

376400

449400

Gebze 1

Turkey

15.00%

33800

35800

68800

135500

133400

184500

230884

Gebze 2

Turkey

15.00%

14000

33000

78000

118000

156300

248200

283903

Gemlik 1

Turkey

15.00%

90500

94800

114500

141000

152300

200500

195021

Gemlik 2

Turkey

15.00%

240500

274600

341300

336300

214100

269300

462987

Gemlik 3

Turkey

15.00%

21800

84700

108100

107322

Aliaga 1

Turkey

15.00%

139918

256598

Aliaga 2

Turkey

15.00%

99414

127961

İzmir

Turkey

14.00%

784400

847900

898200

884900

826600

726700

672486

Thessaloniki

Greece

13.00%

366000

344000

447000

239000

270200

273300

295870

Piraeus 1

Greece

13.00%

1394500

1403400

1373100

433600

498838

178919

490904

Piraeus 2

Greece

12.00%

166062

684881

1188100

Antalya

Turkey

15.00%

11800

40200

63400

67100

59500

125700

165474

Mersin

Turkey

9.00%

594243

632905

799532

869596

845117

1015567

1126866

Limassol

Cyprus

10.00%

320100

358100

377000

417000

353700

348400

345614

Lattakia

Syria

9.00%

390800

472000

546600

568200

621377

586283

524614

Beirut

Lebanon

8.00%

463700

594200

947200

945134

994601

949155

1034249

Haifa

Israel

9.00%

1123000

1070000

1170000

1396000

1140000

1263552

1235000

Ashdod

Israel

11.00%

587000

693000

809000

828000

893000

1015000

1176000

Alexandria 1

Egypt

7.00%

733900

762000

977000

632250

638700

666500

757572

Alexandria 2

Egypt

5.00%

632250

638700

666500

700000

Damietta

Egypt

8.00%

1129600

830100

894200

1125000

1139000

1060100

800000

*Source: Dyamar (2009), Ocean Shipping Consultants (2011) and web pages of related container terminals. ** Market share of

interested feeder shipping line in related container terminal *** Total throughput of related container terminal in TEU

Table A.11: Monthly export and import rates of countries in 2005-2011

Georgia

Year

Jan.

Feb.

Mar.

Apr.

May

Jun.

Jul.

Aug.

Sep.

Oct.

Nov.

Dec.

Export

2005

1.57%

1.45%

2.04%

2.26%

1.72%

2.06%

2.15%

2.34%

2.20%

2.27%

3.25%

2.47%

Import

2005

3.97%

4.16%

5.41%

5.27%

5.23%

5.21%

6.89%

6.67%

6.94%

7.29%

9.03%

8.14%

Export

2006

1.47%

1.31%

1.87%

1.90%

1.61%

1.47%

1.63%

1.67%

1.75%

1.46%

1.92%

2.23%

Import

2006

4.14%

4.85%

5.80%

6.19%

6.38%

6.66%

6.70%

7.88%

7.59%

7.41%

7.23%

8.88%

Export

2007

1.08%

1.05%

1.34%

1.54%

1.75%

1.67%

1.78%

1.72%

1.51%

2.02%

1.72%

1.93%

Import

2007

5.09%

5.19%

5.92%

5.66%

6.80%

5.83%

6.67%

7.19%

6.58%

8.10%

6.91%

10.93%

Export

2008

1.23%

1.30%

1.68%

1.74%

1.85%

2.40%

2.06%

1.44%

2.19%

1.43%

0.91%

0.95%

Import

2008

5.17%

6.08%

6.77%

7.43%

7.83%

7.70%

8.07%

5.62%

6.97%

7.10%

5.68%

6.41%

Export

2009

1.12%

1.35%

1.49%

1.65%

1.78%

1.94%

1.85%

2.04%

1.68%

1.89%

1.77%

2.04%

Import

2009

5.56%

5.52%

6.42%

5.77%

5.75%

6.98%

7.01%

6.65%

6.84%

7.63%

6.93%

8.35%

Export

2010

1.53%

1.58%

1.80%

1.83%

2.08%

1.77%

2.16%

1.75%

2.26%

2.33%

2.37%

2.72%

Import

2010

4.24%

4.78%

6.32%

5.65%

6.40%

6.00%

6.09%

6.55%

6.59%

7.26%

7.13%

8.82%

Export

2011

1.61%

1.59%

1.83%

2.08%

2.16%

1.95%

1.61%

1.99%

1.96%

2.02%

2.14%

2.73%

Import

2011

4.88%

4.58%

6.31%

5.75%

6.07%

6.14%

6.21%

7.38%

6.66%

7.20%

7.22%

7.91%

Romani

Year

Jan.

Feb.

Mar.

Apr.

May

Jun.

Jul.

Aug.

Sep.

Oct.

Nov.

Dec.

Export

2005

2.89%

3.15%

3.69%

3.42%

3.39%

3.56%

3.81%

3.55%

3.95%

3.69%

3.90%

3.55%

Import

2005

3.33%

3.80%

4.58%

4.50%

4.72%

5.01%

4.87%

4.75%

5.02%

5.41%

5.84%

5.64%

Export

2006

2.67%

3.16%

3.51%

2.81%

3.46%

3.42%

3.34%

3.10%

3.32%

3.39%

3.70%

2.94%

Import

2006

3.62%

4.30%

4.94%

4.42%

5.30%

5.23%

5.18%

5.05%

5.08%

5.83%

6.12%

6.10%

Export

2007

2.58%

2.88%

3.27%

2.72%

3.07%

3.12%

3.28%

2.79%

3.10%

3.51%

3.44%

2.81%

Import

2007

4.30%

4.62%

5.28%

4.77%

5.40%

5.34%

5.53%

5.04%

5.21%

6.26%

6.24%

5.45%

Export

2008

2.81%

3.19%

3.06%

3.08%

3.37%

3.42%

3.61%

2.95%

3.29%

3.61%

2.84%

2.15%

Import

2008

4.42%

4.93%

5.37%

5.50%

5.41%

5.74%

5.80%

4.92%

6.03%

5.94%

4.74%

3.82%

Export

2009

2.83%

3.07%

3.81%

3.18%

3.40%

3.77%

4.13%

3.25%

3.84%

4.04%

4.06%

3.44%

Import

2009

3.85%

4.34%

4.79%

4.53%

4.55%

4.89%

4.94%

4.35%

5.52%

5.39%

5.30%

4.75%

Export

2010

2.75%

3.05%

3.60%

3.45%

3.58%

4.00%

4.04%

3.32%

4.19%

4.30%

3.89%

Import

2010

3.32%

3.83%

4.71%

4.50%

4.77%

5.13%

4.82%

4.07%

5.18%

5.08%

5.41%

4.82%

Export

2011

3.42%

3.53%

4.11%

3.41%

3.86%

3.78%

3.80%

3.47%

4.21%

4.12%

4.13%

3.28%

Import

2011

3.64%

3.94%

5.00%

4.40%

5.06%

4.69%

4.49%

4.36%

5.03%

4.91%

5.06%

4.30%

Lubnan

Year

Jan.

Feb.

Mar.

Apr.

May

Jun.

Jul.

Aug.

Sep.

Oct.

Nov.

Dec.

Export

2005

1.24%

1.44%

1.77%

1.98%

2.45%

2.32%

0.89%

0.74%

1.60%

1.53%

1.90%

1.77%

Import

2005

5.49%

6.56%

8.03%

7.26%

8.38%

7.50%

5.14%

2.30%

6.26%

7.28%

8.41%

7.73%

Export

2006

1.24%

1.43%

1.76%

1.97%

2.44%

2.31%

0.89%

0.74%

1.60%

1.52%

1.89%

1.76%

Import

2006

5.50%

6.57%

8.05%

7.27%

8.39%

7.51%

5.14%

2.30%

6.27%

7.29%

8.42%

7.74%

Appendices 147

Export

2007

1.29%

1.50%

1.46%

1.58%

1.59%

1.55%

1.48%

1.52%

1.87%

1.76%

2.00%

1.65%

Import

2007

6.06%

5.68%

6.72%

6.46%

6.35%

6.09%

7.18%

7.12%

6.44%

8.15%

7.21%

7.27%

Export

2008

1.53%

1.82%

1.83%

1.68%

1.61%

1.95%

1.86%

1.87%

1.91%

1.59%

1.88%

1.48%

Import

2008

5.31%

5.68%

6.13%

6.53%

6.10%

5.93%

8.30%

7.03%

6.99%

7.33%

7.98%

5.70%

Export

2009

1.66%

2.12%

1.49%

1.37%

1.67%

1.50%

1.39%

1.46%

1.72%

1.84%

1.87%

2.08%

Import

2009

5.60%

5.19%

5.67%

8.72%

6.08%

7.46%

7.08%

6.84%

6.31%

6.29%

8.01%

6.60%

Export

2010

1.56%

1.72%

1.85%

1.68%

1.79%

1.89%

1.56%

1.61%

1.49%

2.12%

1.67%

2.42%

Import

2010

5.63%

5.43%

8.36%

5.88%

5.93%

6.27%

7.86%

6.51%

5.69%

6.48%

7.90%

6.67%

Export

2011

1.50%

1.41%

1.63%

1.72%

2.61%

2.09%

1.91%

1.76%

1.66%

1.77%

1.61%

1.77%

Import

2011

6.87%

4.73%

6.44%

5.87%

6.23%

5.98%

6.17%

7.10%

7.01%

9.47%

6.23%

6.45%

Cyprus

Year

Jan.

Feb.

Mar.

Apr.

May

Jun.

Jul.

Aug.

Sep.

Oct.

Nov.

Dec.

Export

2005

1.24%

1.29%

1.75%

1.44%

1.56%

1.57%

1.56%

1.45%

1.78%

1.86%

1.91%

1.48%

Import

2005

5.35%

5.88%

7.15%

6.32%

6.66%

6.55%

6.59%

6.00%

7.33%

7.60%

8.04%

7.65%

Export

2006

1.31%

1.29%

1.50%

1.38%

1.34%

1.35%

1.37%

1.14%

1.39%

1.16%

1.66%

1.34%

Import

2006

6.39%

6.38%

6.87%

6.63%

7.34%

8.73%

6.51%

6.83%

7.21%

7.26%

6.78%

Export

2007

1.54%

1.09%

1.25%

1.23%

1.44%

1.18%

1.07%

1.15%

1.36%

1.24%

1.09%

Import

2007

5.86%

5.72%

6.82%

7.26%

7.17%

7.57%

7.45%

7.05%

7.03%

8.03%

7.83%

7.13%

Export

2008

1.09%

1.05%

1.11%

1.13%

0.93%

0.92%

0.88%

1.05%

0.94%

1.11%

1.35%

1.16%

Import

2008

6.91%

5.67%

7.57%

6.69%

7.39%

7.35%

8.76%

7.12%

8.08%

7.77%

7.69%

6.28%

Export

2009

1.01%

1.18%

1.05%

1.37%

1.44%

1.31%

1.30%

0.99%

1.15%

1.35%

1.13%

1.31%

Import

2009

6.60%

6.55%

7.11%

7.64%

7.38%

7.11%

6.94%

7.68%

7.55%

7.05%

7.22%

Export

2010

1.00%

1.04%

1.43%

1.10%

1.28%

1.24%

1.36%

1.07%

1.56%

1.42%

1.48%

Import

2010

6.00%

5.63%

7.60%

6.57%

6.96%

7.43%

6.80%

6.97%

7.27%

6.99%

8.96%

7.38%

Export

2011

1.25%

1.42%

1.65%

1.38%

1.61%

1.63%

1.58%

1.38%

1.48%

1.54%

1.68%

1.50%

Import

2011

6.16%

6.95%

7.46%

6.96%

6.85%

7.07%

6.71%

7.70%

6.04%

6.39%

6.77%

6.82%

Syria

Year

Jan.

Feb.

Mar.

Apr.

May

Jun.

Jul.

Aug.

Sep.

Oct.

Nov.

Dec.

Export

2005

3.01%

3.36%

3.88%

3.65%

3.99%

3.63%

3.91%

4.10%

4.76%

3.98%

3.26%

4.26%

Import

2005

3.62%

3.47%

3.62%

4.33%

4.77%

4.84%

4.56%

5.70%

3.95%

5.52%

3.84%

6.01%

Export

2006

2.85%

3.01%

3.85%

4.31%

5.25%

4.22%

3.88%

4.30%

3.76%

3.07%

3.51%

6.73%

Import

2006

2.62%

3.32%

4.39%

3.84%

4.85%

4.67%

3.99%

3.91%

4.20%

4.76%

4.89%

5.83%

Export

2007

1.91%

2.93%

3.37%

3.99%

4.05%

3.08%

3.17%

3.81%

3.95%

3.55%

5.23%

6.79%

Import

2007

3.99%

4.34%

3.85%

4.08%

5.17%

3.85%

4.53%

6.64%

4.77%

4.15%

4.40%

Export

2008

3.36%

3.88%

4.14%

3.49%

4.22%

4.01%

3.80%

3.77%

4.24%

3.40%

2.81%

4.62%

Import

2008

4.69%

4.22%

5.24%

4.87%

5.22%

4.59%

4.69%

4.96%

3.69%

4.29%

4.41%

3.39%

Export

2009

1.56%

1.93%

2.32%

2.33%

2.98%

3.06%

3.32%

3.38%

3.11%

4.86%

4.47%

7.27%

Import

2009

4.10%

3.83%

4.35%

4.41%

4.29%

4.94%

4.97%

5.77%

4.81%

5.67%

6.57%

Export

2010

2.46%

2.20%

2.95%

3.21%

3.22%

3.66%

2.85%

3.37%

3.31%

4.47%

3.40%

6.11%

Import

2010

4.73%

4.33%

5.04%

4.90%

4.51%

4.60%

4.92%

5.09%

4.65%

5.31%

5.29%

5.43%

Export

2011

2.46%

2.20%

2.95%

3.21%

3.22%

3.66%

2.85%

3.37%

3.31%

4.47%

3.40%

6.11%

Import

2011

4.73%

4.33%

5.04%

4.90%

4.51%

4.60%

4.92%

5.09%

4.65%

5.31%

5.29%

5.43%

Egypt

Year

Jan.

Feb.

Mar.

Apr.

May

Jun.

Jul.

Aug.

Sep.

Oct.

Nov.

Dec.

Export

2005

2.22%

2.24%

2.71%

2.43%

2.71%

2.48%

2.74%

2.82%

3.11%

4.07%

3.53%

3.90%

Import

2005

4.79%

4.52%

5.60%

5.27%

5.99%

5.53%

5.65%

5.49%

5.70%

5.05%

5.53%

5.93%

Export

2006

2.85%

3.54%

3.56%

3.61%

3.56%

3.08%

3.40%

3.02%

3.06%

3.07%

3.66%

3.56%

Import

2006

4.70%

4.48%

4.87%

4.40%

4.93%

4.82%

5.60%

6.23%

4.19%

5.24%

4.86%

5.71%

Export

2007

2.59%

3.26%

3.32%

3.18%

3.25%

2.84%

3.08%

2.64%

2.65%

2.98%

3.88%

3.75%

Import

2007

4.88%

4.42%

5.01%

5.36%

5.10%

4.92%

4.63%

5.53%

4.86%

5.45%

6.23%

6.18%

Export

2008

2.87%

2.79%

3.18%

3.17%

3.07%

3.71%

3.30%

2.40%

2.43%

2.36%

2.12%

1.80%

Import

2008

4.90%

4.65%

5.41%

5.45%

6.03%

5.52%

6.04%

6.79%

5.72%

6.14%

5.55%

4.59%

Export

2009

2.26%

2.63%

2.67%

2.78%

2.83%

3.15%

2.62%

2.82%

2.65%

2.76%

2.86%

3.93%

Import

2009

5.25%

5.24%

5.11%

5.06%

5.05%

5.13%

6.16%

6.43%

4.96%

6.00%

5.15%

6.53%

Export

2010

2.49%

2.54%

2.96%

2.81%

3.11%

2.98%

2.78%

2.68%

2.73%

2.86%

2.75%

3.33%

Import

2010

4.96%

4.39%

5.45%

4.97%

5.17%

5.05%

6.22%

5.95%

5.14%

6.64%

5.66%

6.39%

Export

2011

2.29%

2.56%

3.10%

3.08%

3.22%

3.30%

2.98%

2.49%

2.63%

2.74%

2.75%

3.02%

Import

2011

5.20%

3.67%

5.21%

5.03%

6.38%

5.43%

5.51%

5.89%

6.11%

6.34%

5.07%

5.98%

Ukraine

Year

Jan.

Feb.

Mar.

Apr.

May

Jun.

Jul.

Aug.

Sep.

Oct.

Nov.

Dec.

Export

2005

3.53%

3.75%

4.58%

4.31%

3.96%

3.92%

3.87%

3.94%

4.02%

4.16%

4.17%

4.47%

Import

2005

2.56%

3.42%

4.66%

4.39%

3.96%

4.51%

4.25%

4.64%

4.55%

4.62%

5.20%

Export

2006

2.80%

3.02%

3.75%

3.54%

3.72%

3.98%

4.01%

4.21%

4.41%

4.12%

4.01%

4.43%

Import

2006

3.25%

3.80%

4.64%

3.94%

4.36%

4.33%

4.42%

4.60%

4.97%

4.82%

4.70%

6.16%

Export

2007

2.92%

3.10%

3.74%

3.70%

3.71%

3.85%

3.87%

3.79%

3.74%

3.95%

4.05%

4.37%

Import

2007

3.37%

3.91%

4.51%

4.39%

4.41%

4.26%

4.84%

4.43%

4.41%

5.34%

5.30%

6.03%

Appendices 148

Export

2008

2.40%

3.07%

3.57%

3.65%

4.12%

4.52%

4.99%

4.41%

4.38%

3.84%

2.38%

2.56%

Import

2008

3.03%

4.24%

5.06%

5.20%

5.06%

5.20%

5.79%

5.35%

5.56%

5.01%

3.45%

3.13%

Export

2009

2.87%

3.16%

3.76%

3.63%

3.44%

3.49%

3.77%

4.38%

4.90%

4.64%

4.82%

Import

2009

2.40%

4.46%

4.63%

4.22%

3.76%

4.58%

4.50%

4.77%

5.09%

5.30%

5.89%

Export

2010

2.68%

3.01%

3.52%

3.75%

3.74%

3.86%

3.78%

3.79%

4.19%

4.23%

4.57%

4.72%

Import

2010

2.91%

3.31%

4.21%

4.10%

3.93%

4.21%

4.60%

4.84%

5.06%

5.51%

5.56%

5.93%

Export

2011

3.06%

3.56%

3.71%

3.95%

3.90%

3.55%

3.82%

3.96%

3.79%

4.16%

4.28%

Import

2011

3.34%

4.28%

4.65%

4.17%

4.48%

4.32%

4.77%

4.91%

5.00%

5.08%

5.23%

Bulgaria

Year

Jan.

Feb.

Mar.

Apr.

May

Jun.

Jul.

Aug.

Sep.

Oct.

Nov.

Dec.

Export

2005

2.65%

2.69%

3.28%

3.10%

3.05%

3.40%

3.56%

3.24%

3.20%

3.79%

3.69%

3.56%

Import

2005

3.76%

3.86%

4.65%

4.61%

5.08%

5.35%

5.26%

5.28%

5.20%

5.91%

6.05%

5.78%

Export

2006

2.69%

2.88%

3.20%

3.26%

3.23%

3.52%

3.45%

3.48%

3.44%

3.20%

Import

2006

4.05%

4.01%

4.85%

4.63%

5.00%

4.90%

5.29%

5.51%

5.06%

5.71%

5.58%

6.03%

Export

2007

2.46%

2.56%

3.18%

2.94%

3.15%

3.36%

3.51%

3.24%

3.39%

3.69%

3.58%

3.14%

Import

2007

4.31%

4.07%

4.87%

4.58%

5.01%

5.14%

5.48%

5.16%

5.36%

5.98%

6.15%

5.71%

Export

2008

2.76%

3.01%

3.29%

3.41%

3.28%

3.47%

3.78%

3.27%

3.26%

3.21%

2.66%

2.32%

Import

2008

4.52%

4.73%

4.95%

5.59%

5.56%

6.04%

5.95%

4.92%

5.26%

5.88%

4.75%

4.11%

Export

2009

2.85%

3.15%

3.30%

2.91%

3.17%

3.50%

3.60%

3.44%

3.66%

4.20%

3.80%

3.37%

Import

2009

4.30%

4.64%

5.27%

4.85%

5.01%

5.02%

4.62%

4.92%

5.48%

4.98%

Export

2010

2.62%

2.87%

3.22%

3.37%

3.48%

4.06%

4.39%

4.17%

4.12%

4.20%

4.18%

4.01%

Import

2010

3.34%

3.38%

4.37%

4.60%

4.72%

4.79%

4.82%

4.42%

4.69%

5.06%

5.76%

5.34%

Export

2011

3.58%

3.56%

3.87%

3.58%

3.94%

3.92%

4.39%

4.28%

4.46%

4.36%

3.56%

Import

2011

3.78%

3.65%

4.40%

4.25%

4.46%

3.93%

4.90%

3.91%

4.33%

4.78%

5.16%

4.66%

Greece

Year

Jan.

Feb.

Mar.

Apr.

May

Jun.

Jul.

Aug.

Sep.

Oct.

Nov.

Dec.

Export

2005

1.79%

2.10%

2.05%

2.06%

1.90%

2.00%

2.05%

2.11%

1.95%

2.07%

2.15%

Import

2005

6.36%

6.50%

6.76%

6.58%

5.92%

5.76%

5.92%

6.39%

6.61%

6.24%

6.33%

6.48%

Export

2006

1.96%

1.83%

1.93%

1.76%

2.26%

2.14%

1.96%

2.23%

2.32%

1.98%

2.23%

2.09%

Import

2006

5.78%

5.59%

5.79%

5.95%

6.56%

6.41%

6.22%

6.58%

6.60%

6.47%

6.60%

6.77%

Export

2007

1.89%

1.79%

2.01%

1.81%

1.70%

1.81%

1.85%

2.00%

2.14%

2.07%

2.24%

1.88%

Import

2007

5.76%

5.95%

5.90%

5.92%

6.25%

6.05%

6.50%

6.48%

6.62%

6.82%

7.25%

7.28%

Export

2008

1.74%

1.90%

1.77%

1.95%

2.06%

2.04%

2.13%

1.88%

2.02%

1.89%

1.48%

1.45%

Import

2008

6.34%

6.42%

6.85%

7.25%

7.06%

7.41%

7.23%

6.62%

6.29%

5.94%

5.05%

5.21%

Export

2009

1.63%

1.86%

1.65%

1.78%

1.98%

2.00%

1.95%

1.86%

2.04%

2.06%

1.90%

1.99%

Import

2009

6.32%

6.15%

6.03%

5.91%

6.16%

6.66%

6.67%

6.49%

6.80%

6.43%

6.99%

6.69%

Export

2010

1.93%

1.83%

2.10%

2.02%

1.88%

1.90%

1.86%

2.16%

1.74%

2.60%

2.68%

2.76%

Import

2010

6.97%

6.51%

7.05%

6.15%

5.86%

5.66%

5.80%

5.72%

6.02%

6.51%

6.22%

6.06%

Export

2011

2.50%

2.43%

2.69%

3.12%

3.28%

3.01%

3.06%

3.28%

3.00%

2.78%

2.58%

2.78%

Import

2011

5.96%

5.89%

5.86%

6.19%

6.01%

5.95%

5.89%

5.99%

5.34%

3.98%

4.20%

4.23%

Israel

Year

Jan.

Feb.

Mar.

Apr.

May

Jun.

Jul.

Aug.

Sep.

Oct.

Nov.

Dec.

Export

2005

3.91%

3.59%

4.23%

3.30%

4.16%

3.80%

3.65%

3.68%

4.25%

2.88%

4.04%

3.67%

Import

2005

4.31%

4.21%

4.92%

3.84%

5.34%

4.51%

4.66%

4.82%

5.06%

4.03%

4.48%

4.66%

Export

2006

3.45%

3.46%

4.13%

3.14%

4.60%

3.99%

3.48%

3.56%

3.91%

3.80%

4.17%

3.93%

Import

2006

4.49%

4.20%

4.33%

4.07%

4.92%

4.53%

4.31%

4.78%

4.15%

5.15%

4.52%

4.92%

Export

2007

3.61%

3.25%

3.99%

3.05%

4.32%

3.62%

3.84%

3.41%

3.45%

3.92%

4.44%

4.12%

Import

2007

3.94%

3.98%

4.17%

4.02%

4.55%

4.48%

5.21%

4.99%

4.09%

5.05%

4.88%

5.63%

Export

2008

3.74%

3.71%

4.18%

3.50%

4.29%

4.24%

4.35%

3.44%

3.91%

2.79%

3.25%

2.92%

Import

2008

4.46%

4.40%

5.19%

4.72%

4.99%

5.10%

5.48%

4.94%

4.62%

4.18%

4.02%

3.59%

Export

2009

3.41%

3.17%

4.00%

2.91%

3.99%

3.93%

4.05%

3.66%

4.19%

4.60%

4.42%

4.94%

Import

2009

3.83%

3.72%

4.24%

3.64%

3.86%

4.40%

4.72%

4.97%

4.33%

4.52%

5.17%

5.34%

Export

2010

3.76%

3.38%

4.47%

3.73%

3.85%

4.33%

3.99%

3.55%

3.45%

3.79%

3.98%

4.14%

Import

2010

4.10%

3.85%

4.68%

4.15%

4.38%

4.48%

4.42%

4.79%

3.82%

5.26%

4.47%

5.16%

Export

2011

3.36%

3.47%

4.47%

3.25%

4.23%

4.06%

3.81%

3.60%

3.74%

3.03%

3.75%

3.64%

Import

2011

4.21%

4.13%

5.16%

4.42%

5.19%

4.70%

5.07%

4.83%

4.27%

4.45%

4.78%

4.40%

Turkey

Year

Jan.

Feb.

Mar.

Apr.

May

Jun.

Jul.

Aug.

Sep.

Oct.

Nov.

Dec.

Export

2005

3.10%

3.38%

3.25%

3.24%

3.13%

3.12%

3.01%

3.15%

3.40%

3.36%

3.07%

3.39%

Import

2005

4.68%

5.27%

5.11%

5.07%

4.98%

4.87%

4.86%

5.33%

5.25%

5.37%

5.12%

5.49%

Export

2006

2.75%

3.00%

3.11%

2.98%

3.01%

3.32%

3.15%

3.24%

3.25%

2.96%

3.69%

3.56%

Import

2006

4.52%

5.13%

5.02%

5.23%

5.28%

5.04%

5.03%

5.18%

5.31%

5.11%

5.69%

5.44%

Export

2007

2.82%

3.02%

2.95%

3.14%

3.12%

3.21%

3.39%

3.30%

3.44%

3.82%

3.25%

Import

2007

4.75%

4.79%

4.53%

4.83%

4.95%

4.77%

5.22%

5.07%

5.24%

5.78%

5.99%

5.49%

Export

2008

3.70%

3.46%

3.25%

3.47%

3.50%

3.57%

3.63%

3.62%

3.84%

2.79%

2.76%

2.12%

Import

2008

5.79%

5.39%

4.89%

5.39%

5.38%

5.57%

5.60%

5.66%

5.34%

4.32%

3.72%

3.23%

Appendices 149

Export

2009

3.62%

3.76%

3.22%

3.15%

3.01%

3.35%

3.61%

3.53%

3.64%

3.84%

3.67%

3.73%

Import

2009

4.45%

4.29%

4.20%

4.13%

4.38%

4.80%

4.93%

5.21%

5.17%

5.14%

5.56%

5.62%

Export

2010

3.02%

2.98%

3.10%

3.11%

3.27%

3.18%

2.98%

3.13%

3.11%

3.54%

3.06%

3.58%

Import

2010

4.76%

4.52%

4.80%

4.95%

4.81%

4.75%

4.87%

5.11%

5.15%

5.87%

6.02%

6.32%

Export

2011

2.90%

2.87%

2.88%

3.15%

2.93%

2.96%

3.20%

3.22%

2.95%

3.03%

2.95%

2.92%

Import

2011

5.47%

5.34%

5.26%

5.62%

5.52%

5.23%

5.45%

5.14%

5.50%

5.40%

5.19%

4.94%

Russia

Year

Jan.

Feb.

Mar.

Apr.

May

Jun.

Jul.

Aug.

Sep.

Oct.

Nov.

Dec.

Export

2005

4.52%

4.93%

5.21%

5.49%

5.34%

5.36%

5.84%

5.53%

5.71%

5.86%

6.06%

6.28%

Import

2005

2.55%

2.63%

2.69%

2.72%

2.70%

2.75%

2.88%

2.81%

2.91%

2.98%

3.11%

3.15%

Export

2006

5.05%

5.31%

5.18%

5.40%

5.60%

5.43%

5.48%

5.63%

5.55%

4.93%

5.44%

6.11%

Import

2006

2.43%

2.51%

2.59%

2.61%

2.81%

3.09%

2.85%

2.98%

3.09%

3.11%

3.22%

3.59%

Export

2007

4.28%

4.73%

4.78%

4.80%

4.94%

4.89%

4.86%

4.96%

4.97%

5.61%

6.10%

6.54%

Import

2007

2.81%

2.90%

3.04%

3.01%

3.16%

3.26%

3.20%

3.28%

3.24%

3.39%

3.61%

3.67%

Export

2008

5.18%

5.22%

5.60%

5.53%

5.52%

5.71%

5.77%

5.66%

5.23%

4.68%

4.14%

3.53%

Import

2008

2.97%

3.17%

3.25%

3.45%

3.41%

3.26%

3.59%

3.47%

3.30%

3.06%

2.73%

2.57%

Export

2009

4.32%

4.47%

4.35%

4.49%

4.69%

4.80%

4.98%

5.19%

5.43%

5.75%

6.13%

6.56%

Import

2009

3.22%

3.33%

3.12%

3.11%

3.02%

3.03%

3.01%

3.02%

3.26%

3.44%

3.60%

3.69%

Export

2010

5.22%

5.59%

5.48%

5.31%

5.02%

4.94%

4.65%

4.36%

4.87%

5.08%

5.26%

6.19%

Import

2010

2.70%

2.94%

3.06%

3.07%

3.19%

2.96%

3.13%

3.43%

3.28%

3.35%

3.45%

3.48%

Export

2011

4.36%

5.47%

5.37%

5.89%

5.01%

5.09%

4.86%

4.67%

4.61%

5.10%

5.49%

5.82%

Import

2011

2.94%

3.17%

3.43%

3.40%

3.45%

3.19%

3.15%

3.25%

2.91%

3.01%

3.27%

3.10%

Table A.12: Forecasted weekly total throughputs of regional container terminals*

Burgas

Varna

Constanta 1

Constanta 2

Illiychevsk

Odessa

Weeks

Export

Import

Export

Import

Export

Import

Export

Import

Export

Import

Export

Import

187

203

1002

1150

1271

3011

2913

3068

2195

3151

3824

5079

213

179

1003

1203

1411

2786

2784

2982

2017

2911

3991

3903

218

190

1162

1359

1402

2792

2418

2894

2287

2616

4352

5499

260

200

1031

1291

1165

2861

2680

2567

2415

2662

4069

4975

219

217

885

1642

2827

2426

3555

2669

3095

3880

4963

215

243

877

1736

1820

3389

2683

3027

3053

3304

4771

6326

199

237

830

1908

1662

2988

2874

3612

2901

3406

5012

5590

225

208

820

1683

1829

2921

2795

3204

2892

3267

5155

6578

211

248

828

1310

1447

3173

2416

3297

2775

3339

4969

6365

248

270

784

1342

1425

2823

2688

3497

2311

3061

4657

6455

257

268

875

1331

1298

2890

2977

3598

2351

2861

4156

5804

277

283

890

1369

1508

2794

3078

3699

2182

3164

4824

5841

261

233

777

1543

1694

2838

2964

3262

2660

3173

4304

7082

314

251

722

997

2254

3058

2841

3407

3177

3602

5498

6590

294

231

843

1036

2253

3317

3177

3410

3066

3430

5119

6840

259

242

661

1012

1825

3145

3150

2937

3006

3051

5208

6557

318

242

782

963

2354

3443

2866

3665

2937

3135

5641

7004

304

244

663

936

1957

2930

2885

3412

3053

3254

5578

6353

280

222

809

1105

2321

3091

2814

3349

2833

3513

5221

6002

261

245

785

1038

2087

3328

3091

3296

3084

2736

5660

7084

285

214

717

898

1904

3013

2893

3300

3073

3467

4516

6658

327

231

823

939

1918

3087

2465

2881

2584

3100

5364

6611

460

244

1081

865

1458

2985

2403

4330

2744

3524

5432

6988

405

240

1071

1007

1634

3232

2149

3469

2943

3469

5353

6774

406

241

1056

974

1370

3142

2417

3351

2470

3447

6091

7063

394

300

1212

916

1662

3041

2717

3439

2538

3419

5513

8134

405

421

1397

1235

3935

2987

3264

3414

2540

3660

5205

8092

462

418

1405

1534

3700

3089

3372

3414

2403

3374

5530

7854

455

395

1323

1358

3278

2937

3257

2894

2526

3552

5653

7432

473

351

1432

1595

3193

2848

2594

3174

2734

3493

4795

7343

464

246

1357

1415

2281

3206

2371

3264

2979

4257

4580

7081

460

205

1559

1368

1439

2869

2285

2943

3591

4003

5398

7966

437

194

1452

1216

1595

3106

2618

3670

3390

3943

5619

8387

464

177

1651

1370

1636

3156

2774

3191

3503

3577

5901

8359

390

216

1469

1411

2006

3177

2872

3100

3115

3337

5826

7993

536

257

1499

1238

2270

2906

3116

3511

2826

4574

7506

9296

Appendices 150

448

231

1621

1156

2539

3580

3352

4223

3200

4690

6322

8839

454

261

1600

1299

2110

2909

3328

4965

2648

4491

5319

8439

482

246

1739

1281

2366

3704

3345

3755

2690

4491

6290

8093

410

226

1096

1041

6390

3737

2674

3961

3039

4900

5761

7817

374

224

1104

1237

7444

3238

3171

3560

3120

5075

5834

8422

382

260

1234

1029

7468

3208

2810

4262

2740

4972

5839

8889

366

248

1198

1172

6737

3177

2613

4109

2974

4744

5712

7441

431

336

1346

1315

8217

3352

2691

4558

3275

5483

6312

9106

437

468

1376

1484

10710

3138

3344

4246

4077

6131

6114

7152

495

428

1536

1436

9979

3568

3202

4056

3771

4926

5583

8224

420

422

1583

1689

9530

3553

3646

4331

3587

6148

5163

7485

398

340

1297

1378

7954

3520

2971

3823

3155

6043

5718

6313

358

195

1369

1028

4100

3136

2606

3800

2561

6698

4760

7349

335

187

1357

1006

4211

3210

2375

2906

2328

5952

5197

6703

336

196

1389

1005

4296

2869

2391

3232

2388

6756

5297

6676

382

213

1414

909

3877

2921

2330

3296

2131

6099

5060

6272

Novorossiysk 1

Novorossiysk 2

Poti

Batumi

Trabzon

Haydarpasa

Weeks

Export

Import

Export

Import

Export

Import

Export

Import

Export

Import

Export

Import

3998

2658

2595

1569

905

3911

275

297

286

571

1221

2708

3472

2788

2700

1562

889

4275

303

328

286

484

1403

2660

3949

2707

2350

1361

933

4715

278

289

238

521

1399

2899

4106

2664

2982

1366

931

4662

294

305

258

460

1268

2163

3918

2951

2729

1438

1296

5057

333

346

289

482

1620

2483

3566

2969

3095

1607

1339

4693

304

355

320

583

1505

2470

4375

3229

2832

1815

1396

4379

332

329

320

544

1636

2913

4028

3024

2636

1822

1566

4329

308

346

329

486

1546

2694

4218

2688

2620

1577

1334

5070

347

337

307

535

1400

2398

4584

2836

2703

1577

1422

4494

298

306

289

524

1287

1965

4920

2929

2582

1519

1375

4472

288

304

280

511

1496

2178

5028

2974

2330

1504

1682

5069

301

328

287

469

1243

2403

4189

2678

2551

1426

1529

4934

312

321

239

490

1402

2439

5102

2961

3477

1415

1883

5941

337

359

325

475

1497

2511

5228

3131

2751

1768

1708

4760

298

332

318

478

1485

2307

5008

3227

2538

1659

1710

4982

281

339

284

497

1338

2092

4884

2699

2568

1789

1806

5071

340

334

268

529

1440

2478

4455

3124

2685

1587

1800

5794

329

340

275

439

1334

2212

4674

2493

2668

1731

1686

4951

304

315

271

490

1433

2059

5243

3037

2741

1410

1789

4865

280

361

266

484

1407

2641

5247

3238

2968

1363

1781

4654

300

302

307

499

1308

2420

5450

2864

2794

1664

1641

4930

331

323

293

487

1465

1995

4766

2803

2311

1588

1834

5455

318

296

285

642

1369

2236

4669

2906

2406

1555

1590

5112

328

316

290

558

1332

2112

5021

3117

2504

1405

1681

5043

360

314

344

516

1470

2423

5201

3274

2542

1546

1887

5062

281

309

260

541

1623

2238

4912

2712

3093

1475

1529

5198

313

368

277

514

1289

1896

4814

3074

2702

1387

1465

5226

278

340

285

580

1419

2574

4483

2938

2845

1306

1256

5073

393

336

272

583

1303

2344

4382

3081

2610

1565

1555

5397

324

342

281

524

1423

2031

5447

2924

3065

1787

1590

4981

297

347

298

557

1310

2072

4987

2917

2892

1784

1743

5501

321

353

286

526

1407

2304

4770

3252

2906

1846

1869

4801

321

364

278

497

1593

2129

5130

3247

2865

1592

2012

5343

332

341

298

502

1575

2270

5613

3129

2869

1461

1697

5751

350

360

301

486

1436

1987

5479

3352

3007

1538

1620

4836

358

363

330

544

1553

2269

5367

3051

2867

1840

1716

5884

325

419

323

527

1563

2112

5214

3197

3168

1724

1569

5902

358

375

348

508

1652

2182

5743

3138

3201

1826

1367

5856

357

360

315

543

1626

2239

5614

3397

3621

1677

1711

5405

317

390

331

572

1615

1984

5245

3014

2880

1499

1961

5136

381

440

308

624

1626

2442

5244

3017

3164

1765

1845

6098

378

398

264

600

1709

2243

5318

3454

2957

1496

1940

5749

332

386

313

609

1549

2461

Appendices 151

5008

2645

3033

1803

2105

5829

342

336

334

567

1449

2248

5251

3189

2985

1564

1784

5221

357

318

267

495

1621

2309

4909

3163

2675

1620

1775

4996

357

337

299

515

1446

2359

5386

3308

3175

1759

1966

5664

310

337

285

544

1724

2037

5261

3118

3081

1905

1889

5468

295

344

321

508

1510

2084

4669

2883

2902

1448

1826

4665

301

293

258

519

1554

1837

4861

2835

2923

1502

1733

5404

335

292

303

591

1626

2226

5159

2811

2488

1549

1653

5189

282

334

279

541

1388

1928

4922

2905

2571

1683

1671

4779

331

292

304

531

1294

2032

Ambarli 1

Ambarli 2

Ambarli 3

Gebze 1

Gebze 2

Gemlik 1

Weeks

Export

Import

Export

Import

Export

Import

Export

Import

Export

Import

Export

Import

12558

16359

5605

7766

3107

4921

1688

2939

1624

3594

1443

2658

11777

15446

5341

9240

3464

5750

1616

2923

1750

3629

1705

2436

11780

15432

5730

8871

3143

5219

1476

2856

2157

3696

1455

2486

13682

15345

5585

8142

3290

6027

1644

3169

1755

3410

1373

2306

12185

17018

4926

9164

3239

5453

1725

3346

2265

3831

1343

2664

11742

15372

6243

9409

2899

6392

1714

3130

2181

3900

1539

3032

13348

15687

6687

11747

2702

5557

1967

3003

2322

4050

1510

2887

12096

19454

6534

9274

3466

5643

1802

3040

2140

3345

1439

3077

11707

14980

5281

10179

3787

5872

1518

2967

2203

3471

1559

2441

13492

14190

6148

10034

3851

5429

1335

3331

2083

3475

1591

2728

13866

16121

5844

10627

4859

5158

1509

2587

1824

3822

1710

2822

13233

15001

5105

10073

5181

5234

1513

2830

2077

3587

1742

2456

15910

15577

6176

10343

4109

5854

1546

2881

1853

3301

1791

2807

16092

13239

5674

9746

4182

5744

1632

3069

2160

3243

1848

2635

13853

12774

5309

9653

3930

5401

1571

3528

2028

3926

1729

3167

14822

13282

6015

9706

4036

5588

1429

3130

1999

3323

1713

2908

14619

10066

6669

10431

3645

5021

1771

2937

2194

3174

1643

2662

15201

10931

6004

9673

3507

5114

1652

2949

1938

3250

1805

2812

14212

12351

5838

9982

3452

5336

1493

2905

2165

3118

1553

3039

15488

11736

6788

9639

3684

5566

1586

3092

2243

3640

1655

2625

14689

12406

5077

9662

4299

4846

1644

2896

2008

3220

1887

2709

14508

10119

6474

8466

3890

5697

1604

3149

2026

3746

1719

2290

15284

8711

6187

9814

4353

5425

1571

3519

2173

4483

1876

3104

13662

8382

5521

9381

3692

5700

1678

3500

2265

3627

1819

2947

14749

9229

6613

9361

4345

5858

1789

2717

2060

3601

2044

3308

14977

9018

5844

9591

3736

6176

1830

3277

2302

3989

1730

2953

12360

8587

6448

9431

4608

6129

1799

3408

2183

3836

1711

2362

14369

8878

6076

9860

4168

5552

1669

3295

2051

4174

1775

2835

13569

9747

5646

8824

4299

5911

1514

3115

1875

3735

1785

2290

14460

8668

6051

9148

4302

5905

1864

3216

2052

3997

1620

2503

9620

13404

6119

9752

4040

5563

1554

3063

2218

3639

1479

2591

7980

13071

5800

9432

4494

5795

1825

2547

2403

3700

1518

2656

9257

14049

5993

10575

4405

5956

1894

2523

2053

3556

1448

2623

8068

13872

6276

11071

4330

5871

1787

2584

2211

3969

1539

2963

9785

16045

7289

9363

4568

6016

1651

2500

2040

4138

1324

2876

17157

30450

6241

10683

4228

6192

1724

3679

2366

3575

1244

3124

15553

28010

5639

9656

3818

4818

1549

3369

2548

3755

1262

2970

17054

27267

7031

8699

4213

6879

1506

3458

2045

3982

1324

3095

16443

26597

7501

9347

3956

5750

1561

3123

2181

3448

1135

3045

9976

17972

6175

10723

4871

6410

1659

3476

2297

4044

1443

3311

10011

17162

7025

10495

5355

6696

1546

3967

2080

4745

1410

2862

10025

14726

6776

11169

5315

5723

1710

2999

2489

4080

1196

3175

10789

18919

6826

10236

5133

6100

1542

3499

2456

4522

1513

3108

13969

11589

6420

10518

4773

5871

1768

3405

2154

3770

1746

2582

14039

10982

5805

10617

3868

5992

1799

2855

2007

3350

1939

2583

13680

9482

6919

9892

4254

6338

1902

3766

1967

3633

2067

2902

16770

10115

6694

10152

3784

5056

1767

3681

2410

3568

1738

2788

12524

12133

5720

10531

4662

6726

1751

3254

1995

3667

1906

2904

5577

12040

6062

8942

5135

5658

1876

2466

2065

3867

1811

2218

6238

14473

5827

9034

4914

5405

1822

2671

1981

3904

1774

2233

Appendices 152

6200

13941

5659

9312

5067

5935

1868

2452

2210

4179

1783

2592

6140

12539

6856

9303

4597

5623

1462

2562

2150

3866

1699

2562

Gemlik 2

Gemlik 3

Aliaga 1

Aliaga 2

Izmir

Thessaloniki

Weeks

Export

Import

Export

Import

Export

Import

Export

Import

Export

Import

Export

Import

2697

3038

793

1029

1595

2775

933

1522

4285

8256

1994

3038

2724

3379

747

1161

1879

2668

773

1370

4381

8645

2210

2706

2928

3711

696

1253

1830

3020

848

1650

4194

8422

2292

2611

2694

3069

707

1288

1689

2804

781

1621

3896

8650

1764

2893

2854

2872

818

1365

1698

3407

885

1505

4864

7511

2287

3094

3055

2492

885

1438

2064

3229

1006

1718

4940

9922

1882

3335

2858

2580

1009

1516

1835

3033

906

1643

4991

10303

2089

2709

3215

2212

888

1429

1953

3119

978

1560

5178

9482

1949

3199

3336

3195

764

1352

2067

2891

880

1490

4833

8301

1313

3169

2511

4306

875

1313

1545

3020

899

1568

4441

8648

868

2362

2966

3977

829

1370

2039

2815

859

1686

5051

9922

872

2592

2843

4247

786

1451

1694

2872

871

1519

4693

9769

723

2559

2634

4080

776

1256

1582

2987

799

1564

4565

8992

888

2638

3478

5048

970

1461

1724

2960

884

1447

4259

10555

1785

2598

3219

4278

997

1272

1881

3281

930

1612

4719

10536

1917

2918

3395

5047

988

1382

2089

3333

874

1640

4821

9731

1764

2590

2845

4392

1071

1407

1875

3058

970

1637

4866

11152

2113

2470

3250

4378

913

1309

1876

3313

799

1411

4506

9558

1828

2636

3043

4830

974

1441

1840

3026

960

1582

4797

9418

1899

2624

2793

4856

984

1328

1975

3006

816

1652

4877

11118

1685

2661

3381

4534

907

1299

1777

3175

897

1609

4550

11069

1894

2668

3333

4733

1012

1391

1706

2894

972

1591

4429

9025

1813

2782

3328

6208

639

1520

1893

3436

889

1575

4562

8644

1742

4048

3783

6554

720

1454

2003

2763

889

1678

5259

8558

1593

3379

2831

6484

709

1398

1894

3336

1016

1606

5227

9104

1789

3583

3730

6699

760

1367

1871

3306

922

1631

4575

10001

1748

4079

2813

5048

1049

1390

1928

3158

870

1562

5106

13190

811

3788

2634

4960

940

1398

2025

3275

820

1852

4854

11770

875

3636

2991

4813

997

1412

1873

2999

989

1713

4413

10837

732

4179

2766

5067

1050

1407

1816

3175

939

1704

4829

10502

814

3945

3049

2770

806

1564

1941

3016

920

1441

4981

12610

1790

3928

3566

2030

724

1607

1860

3040

984

1640

4855

11869

2397

4099

2792

2226

755

1277

2011

3275

958

1271

5116

12042

2428

4243

3064

2207

863

1611

1853

3632

1006

1479

4789

11813

2113

4886

2801

3135

842

1759

1949

3003

901

1577

4438

12467

2039

4145

2530

4368

1108

1516

2099

3223

995

1627

5142

10814

1544

3690

2455

3814

1011

1532

2181

3074

965

1755

5119

9760

1493

3271

2531

4363

1157

1553

1892

3542

1066

1803

5686

10445

1510

3936

2202

4149

1060

1535

2081

3815

965

1748

5034

10217

1485

4065

1486

1875

967

1307

2068

3143

866

1499

5264

7177

970

3181

1552

2215

981

1568

1871

3619

1028

1755

4482

8158

894

3168

1711

2144

912

1446

1967

3488

1138

1645

5463

9053

847

3970

1840

2062

927

1597

2130

3431

949

1834

5489

8616

873

3946

1782

2070

812

1484

1973

3622

1041

1828

4915

8303

1698

3680

1987

2150

746

1565

1946

3183

986

1888

4866

10715

2826

3284

1921

2010

707

1377

2004

3409

897

1975

5234

10994

2615

3459

2422

2082

695

1216

2019

2937

944

1675

5147

9570

2797

3906

1975

2034

725

1260

2060

3556

1003

1548

4809

9860

2127

4027

1629

2151

877

1391

1732

3164

961

1317

4918

6901

907

3939

1374

1731

855

1303

1829

3160

805

1361

4824

8978

1220

4363

1488

2076

1024

1194

2008

3095

958

1564

4296

7606

988

4511

1513

2017

1154

1481

1732

2864

900

1534

4427

7953

958

3955

Piraeus 1

Piraeus 2

Antalya

Mersin

Limassol

Lattakia

Weeks

Export

Import

Export

Import

Export

Import

Export

Import

Export

Import

Export

Import

2767

5976

5932

11021

1028

2335

6986

14574

938

5437

3130

5832

2501

6110

5885

14383

1105

2007

7985

15149

942

5007

3536

5148

2668

5874

6011

12887

1078

2150

8154

14749

848

4715

3015

6178

Appendices 153

2818

5174

6246

12988

1185

1869

8169

12677

924

4165

2967

6063

2731

7449

6653

13927

1101

2016

9016

12586

973

6120

3036

6144

3244

9906

6857

14439

1249

1846

8857

15808

1046

7136

2717

5295

2725

8805

6662

14183

1258

2274

9039

14814

949

7332

3408

5965

3041

9494

6673

14821

1158

2058

9928

14837

1048

6109

3167

6147

2857

8725

5893

13799

1193

2339

8394

13745

1071

7035

3487

5657

2380

8331

6783

14768

1168

2000

8413

14580

1114

5693

5281

5427

2852

9451

6310

13221

992

2243

9015

14152

1312

6012

5125

5996

2656

7480

6327

12435

1098

1971

7400

14473

1081

6535

4614

4881

2792

8256

6073

14696

1100

2078

9188

17021

1383

5456

5103

5248

3104

10647

6563

14396

1125

2073

9143

13573

1454

7713

4632

6660

3002

8624

6706

16293

1116

2409

9125

15130

1484

7382

4515

6584

2483

11674

7150

14673

1231

2381

9177

13335

1347

6755

4255

6508

2577

9499

6351

13285

1266

2210

8976

15666

1297

7551

4411

6381

2706

9943

6614

11941

1099

2189

7686

14045

1339

7040

4547

6528

2863

8911

6263

15515

997

2387

9061

15194

1506

6544

4183

5821

2680

10950

6349

14852

1179

2365

8731

14813

1251

7789

4276

6221

3019

10389

7264

13479

1157

1998

8712

12655

1303

6634

4420

6194

2838

9643

6627

16830

1245

2148

10234

14376

1152

7576

4005

6388

2802

10565

6665

13730

1225

2052

9136

14710

1038

6250

5067

7074

2960

9342

6668

17287

1111

2109

8919

15569

831

6715

4900

6717

2896

9246

7008

16091

1155

2007

9350

14633

942

6270

4053

6874

3011

8934

6749

16380

1220

2071

7900

14493

942

6810

4925

8144

3175

11992

7440

17080

1095

2594

9433

16312

1077

8681

5976

5811

3192

11578

7021

16976

1201

2119

8336

16444

940

8339

5214

5223

3403

12670

7999

17932

1218

2188

9936

14364

1022

7007

5038

6301

3362

11830

8051

15457

1152

2427

8896

15415

990

7551

5478

6425

2782

9517

6957

16264

1252

1758

8360

14403

1133

7974

4905

6556

3056

8925

7801

18211

1275

1457

9415

13021

1333

7186

4950

6803

3386

9521

7062

15292

1043

1725

8495

12689

1297

8737

4765

6362

2818

8576

6920

17774

1154

1727

9743

14550

1190

8323

4804

6048

3139

7644

7089

16749

1382

1785

9445

12906

1284

7961

4730

6868

3102

3158

7631

18197

1373

2347

9815

16423

1625

9234

4459

5988

3072

3037

9458

17625

1242

2677

9849

17708

1589

8590

4414

6484

3057

3141

8696

18468

1330

2671

9862

15558

1463

9610

5480

6942

3217

3234

8305

18280

1086

2771

10025

17562

1881

9600

4898

5906

3110

6103

8577

19743

1154

1905

9549

15429

1063

8484

5868

7362

3472

7228

8267

14591

1157

2130

10033

13170

1099

9478

6273

7866

2992

6693

8641

15108

1293

1775

9480

16260

1330

8929

5532

7784

3144

6896

7368

16567

1375

1978

9693

16247

1389

8274

5815

6875

3522

6666

8600

15553

1408

2181

8082

14042

1154

9893

5881

7694

3102

5164

8276

15446

1075

2608

8606

15861

938

9130

5815

7004

2799

5218

8473

15680

1201

2534

9050

17879

1054

8374

5022

6679

2953

6164

8704

15852

1274

2339

10095

16973

934

7893

5761

8334

3158

4086

8710

14897

1257

2114

9557

15083

994

7946

5708

6089

3159

2459

7074

14333

1016

1234

8588

16294

927

8248

6769

5208

2989

2324

8297

14711

1271

1144

9063

12749

1158

8956

6154

5447

2600

2580

6675

15639

1151

1075

8685

13892

1216

9487

7243

5346

2573

2540

7903

13050

1154

1287

9648

14894

1144

8675

6724

5317

Beirut

Haifa

Ashdod

Alexandria 1

Alexandria 2

Damietta

Weeks

Export

Import

Export

Import

Export

Import

Export

Import

Export

Import

Export

Import

4865

7422

8327

11031

7815

7924

3766

12995

3628

7297

5283

11532

4903

8327

8193

11214

7856

7377

3856

14730

4230

7180

4873

11828

5802

9194

7321

11331

7259

7986

4607

13291

4087

6896

5545

13007

5289

9388

7971

10919

8443

7283

4203

11536

3397

6612

4767

13844

4075

13216

9543

12588

10684

10308

4043

12566

4346

7727

6647

11333

2775

13864

9749

15062

13944

12220

4813

11540

3474

7563

7225

12209

3292

14126

8864

15102

11662

10985

4796

10206

4318

7793

8103

11981

3263

15751

9581

13069

14347

11747

5596

11105

3815

8981

6479

11437

3015

15626

10652

15193

10953

12362

5462

13069

4749

9600

9463

13754

2929

18771

12631

17744

13587

13990

6671

13854

4241

10956

8895

14938

Appendices 154

3200

15539

11243

16706

13068

13448

5892

14188

3855

9785

8922

13454

3146

20252

12691

13989

9466

16078

5620

14010

4006

10312

8201

13237

3748

18194

11656

15933

13021

15676

6513

13558

4039

10030

8880

12126

4543

14232

11549

19423

13935

11917

8052

11009

4279

10974

8008

15854

4025

16174

11438

16676

12448

10522

7919

10432

3879

10234

8618

18760

4960

13642

11611

16781

12975

11638

7850

10218

4397

9861

8543

18294

4456

15099

13022

16512

14117

12376

6887

11301

4601

10200

10021

16945

4857

15169

12142

17449

13459

11117

7023

9367

4200

9971

9098

17028

4391

13967

11545

15166

12767

12479

7365

10672

4140

9387

9468

17425

4757

14018

10348

18706

11832

10055

7873

9729

4372

9531

7849

16335

3838

14189

9938

16351

13083

10895

7306

10596

4085

10378

8472

15313

5346

13635

12326

14935

14221

13475

7267

8527

3887

8665

6711

16184

5516

14879

14534

13693

13074

13763

8663

6376

3990

7050

6572

14931

5544

16287

13864

13093

12926

18135

8848

6650

4612

7944

6671

15804

5618

15414

12561

12855

13035

14912

8724

6518

4631

7956

6846

13526

4737

17191

13008

12438

14352

15559

8870

8143

4027

8418

6775

18430

5682

21077

8411

15966

14240

13148

10500

13375

3474

10565

8354

14939

5212

21741

8410

14715

12778

13641

8541

13238

3469

11823

8425

15489

5399

21264

8866

16635

12903

13415

7523

12249

3595

11041

7204

15160

5411

19919

8497

13477

12431

14402

7562

12711

3518

10756

7749

16952

3779

16098

12835

11428

11068

10725

8961

7834

4106

10536

7567

17914

4048

13336

16294

10845

12402

11815

9213

6925

4013

12101

7382

18396

3588

12954

14427

11126

15386

10395

10017

7030

4344

10683

7332

16644

3369

13564

13625

10108

12665

9376

10059

6117

4591

10719

7872

17752

3722

15221

12860

12759

12342

12789

9086

7294

4521

10577

7474

17060

4750

21523

15112

18410

12054

16389

10495

7632

4885

8601

10979

12290

4493

19531

16681

20671

11287

16212

9906

7597

5157

8521

8803

12573

4837

17650

15520

19779

11340

17031

9740

6930

5149

8155

10864

13110

4665

17984

15017

18004

11588

16433

10224

7228

4558

8458

9655

13034

1436

19919

11437

21637

8632

11269

8729

7050

4314

11683

13342

20797

1268

23481

9873

19160

8943

13358

9162

7295

4463

13138

10900

19789

1459

19738

10417

20390

8991

12612

10133

7397

4230

11928

12039

22346

1502

20978

10037

20088

8599

11421

9047

7026

4087

11516

13431

21527

3579

20126

13204

17566

11518

13983

8858

7172

3864

9599

9906

14620

5143

18392

16134

19306

11168

14743

7643

6987

5369

11767

9617

10762

5141

16907

15436

17525

12224

16528

7255

6485

4640

10787

10613

10872

5795

17394

15753

17806

12844

17950

7448

7259

4510

11258

9123

11020

4951

15316

15335

15749

11183

14182

5459

8928

3608

8927

7875

10155

3752

6867

10930

11214

10546

14325

5468

13330

3424

10272

6472

10190

4070

6285

10972

10293

9869

12946

4726

12483

3444

11015

6011

11219

4144

6393

10674

9066

10223

13650

4973

12144

3299

9559

6438

10655

3821

6401

10188

8864

9354

12773

5052

12173

3438

9623

7123

9463

Total throughput of related container terminals, the operation amount of interested feeder shipping line could be calculated by using

market share ratios in Table A.10.

Table A.13: Results of designed experimental tests

Test 1

Scenario A

Scenario B

Scenario C

Scenario D

Result

309855.9

316330.0

320460.0

323732.3

324134.4

340600.3

Test 2

Scenario A

Scenario B

Scenario C

Scenario D

Result

327220.0

347200.0

357500.0

363880.8

377676.9

404373.1

309855.9

Appendices 155

316330.0

320460.0

323732.3

324134.4

340600.3

280241.0

262368.2

253920.7

246371.3

237288.7

218361.3

298832.7

292802.2

283924.0

276514.3

265712.6

258623.5

302509.6

301000.0

301453.8

301978.9

300421.2

304585.8

Test 3

Scenario A

Scenario B

Scenario C

Scenario D

Result

309855.94

316330.00

320460.00

323732.28

324134.35

340600.33

272609.09

278270.00

283941.90

286478.88

284813.16

299669.71

268230.00

277350.00

276907.46

283280.75

283309.51

299669.71

278951.00

283800.00

286003.42

289055.76

285777.16

299669.71

282090.00

284584.69

290578.67

291378.70

291052.98

299669.71

274360.00

280860.00

282710.00

287904.92

287291.76

299669.71

Test 4

Scenario A

Scenario B

Scenario C

Scenario D

Result

Appendices 156

320463.00

283942.00

276907.00

286004.00

290579.00

282708.00

320463.00

284384.00

280418.00

286933.00

292077.00

284638.00

320463.00

285310.00

283817.00

287104.00

291293.00

284666.00

320463.00

285383.00

285187.00

287156.00

290332.00

285799.00

320463.00

286623.00

288002.00

100

287049.00

101

290149.00

102

286410.00

103

320463.00

104

287820.00

105

293124.00

106

286340.00

107

291569.00

108

287750.00

157

References

Agarwal R, Ergun O (2008) Ship scheduling and network design for cargo routing in

liner shipping. Transportation Science 42:175-196

Ai TJ, Kachitvichyanukul V (2009) A particle swarm optimization for the vehicle

routing problem with simultaneous pickup and delivery. Computers & Operations

Research 36:1693 -- 1702

Alphaliner (2010) The containership market in 2009, Alphaliner annual reports.

http://www.alphaliner.com/. Accessed 14.03.2013

Alphaliner (2011) Weekly newsletter 48. http://www.alphaliner.com/. Accessed

14.03.2012

Alphaliner (2013a) Cellular fleet statistics 03/2013. http://www.alphaliner.com/.

Accessed 11.03.2013

Alphaliner (2013b) TOP 100 operated fleets.

http://www.alphaliner.com/top100/index.php. Accessed 18.03.2013

Altınel K, Öncan T (2005) A new enhancement of the Clarke and Wright savings

heuristic for the capacitated vehicle routing problem. Journal of the Operational

Research Society 56:954–961

Alvarez JF (2009) Joint routing and deployment of a ﬂeet of container vessels. Maritime

Economics & Logistics 11:186-208

Andersen MW (2010) Service network nesign and management in liner container

shipping applications (Chapter 5). Ph.D. Thesis, Technical University of Denmark,

Lyngby, Denmark

Baird AJ (2006) Optimising the container transhipment hub location in northern

Europe. Journal of Transport Geography 14:195-214

Bendall HB, Stent AF (2001) A scheduling model for a high speed containership

service: A hub and spoke short-sea application. International Journal of Maritime

Economics 3 (3):262-277

158

Berbeglia G, Cordeau J-F, Gribkovskaia I, Laporte G (2007) Static pickup and delivery

problems: a classification scheme and survey. TOP 15 (1):1-31

Bose NK, Liang P (1996 ) Neural network fundamentals with graphs, algorithms, and

applications. McGraw-Hill, Inc., Hightstown, NJ, USA

Bukša J, Bukša T (2011) Quality management in coastal container liner shipping from

the point of view of cost management. Scientific Journal of Maritime Research 25

(2):387-403

Bukša J, Kos S (2005) Analysis of the structure of transported containers in the Adriatic

feeder service with a retrospect on the proportion and effect of empty containers on

the service. Scientific Journal of Maritime Research 19 (1):89-99

Bukša J, Zec D (2011) Technology-based risk in coastal container liner shipping.

Scientific Journal of Maritime Research 25 (1):125-144

BunkerIndex (2012) South Europe bunker prices.

http://www.bunkerindex.com/prices/seurope.php. Accessed 15.06.2012

BunkerIndex (2013) Bunker Index 380 CST summary.

http://www.bunkerindex.com/prices/bixfree.php?priceindex_id=2. Accessed

19.03.2013

Catalani M (2009) Ship scheduling and routing optimization: An application to Western

Mediterranean area. European Transport 42:67-82

Catay B (2010) A new saving-based ant algorithm for the vehicle routing problem with

simultaneous pickup and delivery. Expert Systems with Applications 37:6809-6817

Chan CT, Lee HH (2000) Containers, containerships & quay cranes: A practical guide.

Genesis Typesetting & Publication Services, Singapore

Chang Y-T, Lee S-Y, Tongzon JL (2008) Port selection factors by shipping lines:

Different perspectives between trunk liners and feeder service providers. Marine

Policy 32:877-885

Chen C, Zeng Q (2010) Designing container shipping network under changing demand

and freight rates. Transport 25 (1):46-57

159

Chen F-e, Zhang R-y (2008) Economic viability of mega-size containership in different

service networks. Journal of Shanghai Jiaotong University 13 (2):221-225

Chen S-H, Chen J-N (2010) Forecasting container throughputs at ports using genetic

programming. Expert Systems with Applications 37:2054-2058

Cheung RK, Chen CY (1998) A two-stage stochastic network model and solutions

methods for the dynamic empty container allocation problem. Transportation

Science 32 (2):142-162

Chou M, Song M, Teo C (2003) Inventory-routing problem in sea freight: direct versus

transshipment model. Technical Report. The Logistics Institute - Asia Pacific,

Christiansen M, Fagerholt K, Nygreen B, Ronen D (2007) Maritime transportation. In:

C. Barnhart GL (ed) Transportation. Handbooks in Operations Research and

Management Science, vol 14. North-Holland, Amsterdam, pp 189-284

Christiansen M, Fagerholt K, Nygreen B, Ronen D (2013) Ship routing and scheduling

in the new millennium. European Journal of Operational Research 228 (3):467–483

Christiansen M, Fagerholt K, Ronen D (2004) Ship routing and scheduling: Status and

perspective. Transportation Science 38 (1):1-18

Christoﬁdes N, Mingozzi A, Toth P (1979) The vehicle routing problem. In:

Christoﬁdes AM, P. Toth, & C. Sandi (ed) Combinatorial optimization. Wiley,

Chichester, pp 315-338

Chuang T-N, Lin C-T, Kung J-Y, Lin M-D (2010) Planning the route of container ships:

A fuzzy genetic approach. Expert Systems with Applications 37:2948-2956

Clarke G, Wright JW (1964) Scheduling of vehicles from a central depot to a number of

delivery points. Operations Research 12:568-581

Croes GA (1958) A method for solving traveling salesman problems. Operations

Research 6:791-812

Cullinane K, Khanna, M and Song, D-W. (1999) How big is beautiful: Economies of

scale and the optimal size of containership. Paper presented at the Proceedings of

International Association of Maritime Economists, Halifax,

160

de Gooijer JG, Klein A (1989) Forecasting the Antwerp maritime steel traffic flow: a

case study. Journal of Forecasting 8:381-398

Dethloff J (2001) Vehicle routing and reverse logistics: the vehicle routing problem

with simultaneous delivery and pick-up. OR Spektrum 23:79-96

Dong J-X, Song D-P (2009) Container ﬂeet sizing and empty repositioning in liner

shipping systems. Transportation Research Part E 45:860-877

Dong J-X, Song D-P (2012) Quantifying the impact of inland transport times on

container fleet sizing in liner shipping services with uncertainties. OR Spectrum 34

(1):155-180

Doyuran T, Çatay B (2011) A robust enhancement to the Clarke-Wright savings

algorithm. Journal of The Operational Research Society 62 (1):223-231

Ducruet C, Notteboom T (2010) The worldwide maritime network of container

shipping: Spatial structure and regional dynamics. GaWC Research Bulletin 364

(halshs-00538051-v1)

Ducruet C, Notteboom T (2012) Developing liner service networks in container

shipping. In: Song DW, Panayides P (eds) Maritime Logistics: A complete guide to

effective shipping and port management. Kogan Page, Londan, pp 77-110

Dyamar (2009) Black Sea container trades, ports & terminals, hinterland. Consultancy

Reports. Dyamar B.V., Alkmaar

Evers J, Feijter Rd (2004) Centralized versus distributed feeder ship service: the case of

the Maasvlakte harbour area of Rotterdam. Transportation Planning and Technology

27 (5):367-384

Fagerholt K (1999) Optimal ﬂeet design in a ship routing problem. International

Transactions in Operations Research 6 (5):453-464

Fagerholt K (2004) Designing optimal routes in a liner shipping problem. Maritime

Policy & Management 31 (4):259-268

161

Fard MK, Akbari MR (2013) A hybrid tabu search algorithm for the vehicle routing

problem with simultaneous pickup and delivery and maximum tour time length.

African Journal of Business Management 7 (11):801-810

Feng C-M, Chang C-H (2008) Empty container reposition planning for intra-Asia liner

shipping. Maritime Policy & Management 35 (5):469-489

Foschi AD (2003) The maritime container transport structure in the Mediterranean and

Italy. University of Pisa, Department of Economics Paper No 2003-24:1-33

Francesco MD, Crainic TG, Zuddas P (2009) The effect of multi-scenario policies on

empty container repositioning. Transportation Research Part E: Logistics and

Transportation Review 45 (5): 758-770

Francesco MD, Lai M, Zuddas P (2013) Maritime repositioning of empty containers

under uncertain port disruptions. Computers & Industrial Engineering 64 (3):827-

837

Francesetti DC, Foschi AD (2002) The impact of hub and spokes networks in the

Mediterranean peculiarity. Paper presented at the Annual Conference of the

International Association of Maritime Economists, Panama/Panama,

Frankel EG (2002) The challenge of container transshipment in the Caribbean. Paper

presented at the International Association of Maritime Economist Conference,

Panama,

Fung MK (2001) Competition between the ports of Hong Kong and Singapore: a

structural vector error correction model to forecast the demand for container

handling services. Maritime Policy & Management 28 (1):3- 22

Gajpal Y, Abad P (2009) An ant colony system (ACS) for vehicle routing problem with

simultaneous delivery and pickup. Computers & Operations Research 36 (12):3215-

3223

Gelareh S, Meng Q (2010) A novel modeling approach for the ﬂeet deployment

problem within a short-term planning horizon. Transportation Research Part E

46:76-89

162

Gelareh S, Nickel S (2011) Hub location problems in transportation networks.

Transportation Research Part E 47:1092-1111

Gelareh S, Nickel S, Pisinger D (2010) Liner shipping hub network design in a

competitive environment. Transportation Research Part E 46:991-1004

Gelareh S, Pisinger D (2011) Fleet deployment, network design and hub location of

liner shipping companies. Transportation Research Part E 47:947-964

Goksal FP, Karaoglan I, Altiparmak F (2012) A hybrid discrete particle swarm

optimization for vehiclerouting problem with simultaneous pickup and delivery.

Computers & Industrial Engineering doi:10.1016/j.cie.2012.01.005

Gosasang V, Chandraprakaikul W, Kiattisin S (2011) A comparison of traditional and

neural networks forecasting techniques for container throughput at Bangkok port.

The Asian Journal of Shipping and Logistics 27 (3):463-482

Goulielmos AM, Kaselimi E (2011) A non-linear forecasting of container traffic: the

case-study of the Port of Piraeus, 1973-2008. International Journal of Shipping and

Transport Logistics 3 (1):72 - 99

Goulielmos AM, Pardali AI (2002) Container ports in Mediterranean sea: A supply and

demand analysis in the age of globalization. International Journal of Transport

Economics 29 (1):91-117

Guo Z, Song X, Ye J (2005) A verhulst model on time series error corrected for port

throughput forecasting. Journal of the Eastern Asia Society for Transportation

Studies 6:881 - 891

Halse K (1992) Modeling and solving complex vehicle routing problems. Department

of Mathematical Modeling, Technical University of Denmark.,

Ham JCv, Autekie P (2005) Container terminals in the Mediterranean region. Paper

presented at the International Conference on Maritime Engineering, Ports and

Waterways, Barcelona, Spain,

Hansen P, Mladenović N (2001) Variable neighborhood search: principles and

applications. European Journal of Operational Research 130:449-467

163

Hansen P, Mladenović N, Pérez JM (2010) Developments of variable neighborhood

search. Annals of Operations Research 175 (1):367-407

Hemmelmayr VC, Doerner KF, Hartl RF (2009) A variable neighborhood search

heuristic for periodic routing problems. European Journal of Operational Research

195 (3):791-802

Hlača B, Babić AO (2006) Feeder service- logistic support to the port of Rijeka. Promet

- Traffic and Transportation 18 (3)

Hoff A, Andersson H, Christiansen M, Hasle G, Løkettangen A (2010) Industrial

aspects and literature survey: Fleet composition and routing. Computers &

Opertions Research 37 (12):2041-2061

Hsu C-I, Hsieh Y-P (2005) Direct versus terminal routing on a maritıme hub-and-spoke

container network. Journal of Marine Science and Technology 13 (3):209-217

Hsu C-I, Hsieh Y-P (2007) Routing, ship size, and sailing frequency decision-making

for a maritime hub-and-spoke container network. Mathematical and Computer

Modelling 45:899-916

Hui ECM, Seabrooke W, Wong GKC (2004) Forecasting cargo throughput for the Port

of Hong Kong: Error correction model approach. Journal of Urban Planning and

Development 130 (4):195-203

Hwang H-S, Bae S-T, Cho G-S (2007) Container terminal demand forecasting

framework using fuzzy-GMDH and neural network method. Paper presented at the

International Conference on Innovative Computing, Informatio and Control,

Kumamoto, Japan,

Iannone F, Thore SA, Forte E (2007) Inland container logistics and interports. Goals

and features of an ongoing applied research. Paper presented at the The Ninth

Scientific Meeting of the Italian Society of Transport Economists, Naples, Italy.,

October 3-5, 2007

IHS (2012) IHS Global Insight, World Trade Service Database.

Imai A, Shintani K, Papadimitriou S (2009) Multi-port vs. Hub-and-Spoke port calls by

containerships. Transportation Research Part E 45 (5):740-757

164

ISL (2012) Shipping statistics and market review. Statistical Publications, vol 56-5/6.

Institute of Shipping Economics and Logistics, Bremen

Jadrijević N, Tomašević M (2011) Feeder service development in Mediterranean ports

toward more efficient transport technologies. Paper presented at the 14th

International Conference on Transport Science - ICTS Portoroz, Slovenia,

Jansson JO, Shneerson D (1982) The design of liner shipping serives: the problem of

feeder services versus multi-port-calls. Maritime Policy & Management 9 (3):175-

188

Jin Z-H, You Q-H, Wang J (2005) The container ship routing problem and its practical

solutions. Paper presented at the International Conference on Management Science

and Engineering Incheon/Korea

Jun Y, Kim B-I (2012) New best solutions to VRPSPD benchmark problems by a

perturbation based algorithm. Expert Systems with Applications 39 (5):5641-5648

Karlaftis MG, Kepaptsoglou K, Sambracos E (2009) Containership routing with time

deadlines and simultaneous deliveries and pick-ups. Transportation Research Part E

45 (1):210-221

Khashei M, Bijari M, Ardali GAR (2009) Improvement of auto-regressive integrated

moving average models using fuzzy logic and artificial neural networks (ANNs).

Neurocomputing 72:956- 967

Kjeldsen KH (2011) Classification of ship routing and scheduling problems in liner

shipping. INFOR: Information Systems and Operational Research 49 (2):139-152

Kjeldsen KH, Lysgaard J (2011) Routing and scheduling in liner shipping: A column

generation based heuristic for routing ships and cargo in liner shipping. PhD Thesis,

Aarhus University, Aarhus, Denmark

Kline DM, Zhang GP (2004) Methods for multi-step time series forecasting with neural

networks. In: Zhang GP (ed) Neural Networks for Business Forecasting. Information

Science Publishing, pp 226-250

Kolar V, Marinović P, Luburić G (2000) Container- Feeder Service Rijeka (Ploce)-

Gioia Tauro (Malta). Promet - Traffic and Transportation 12 (4):185-188

165

Kulak O, Polat O, Gujjula R, Günther H-O (2013) Strategies for improving a long-

established terminal’s performance: a simulation study of a Turkish container

terminal. Flexible Services and Manufacturing Journal DOI:10.1007/s10696-011-

9128-x

Kulak O, Sahin Y, Taner ME (2011) Joint order batching and picker routing in single

and multiple-cross-aisle warehouses using cluster-based tabu search algorithms

Flexible Services and Manufacturing Journal 24:52-80

Kytöjokia J, Nuortiob T, Bräysya O, Gendreauc M (2007) An efficient variable

neighborhood search heuristic for very large scale vehicle routing problems.

Computers & Operations Research 34 (9):2743-2757

Lachner SA, Boskamp VAG (2011) Routing and scheduling in liner shipping with

multi-start local search heuristics. Master Thesis, Erasmus University, Rotterdam,

the Netherlands

Lam JSL (2010) An integrated approach for port selection, ship scheduling and ﬁnancial

analysis. Netnomics 11:33-46

Lam JSL, Yap WY (2011) Dynamics of liner shipping network and port connectivity in

supply chain systems: analysis on East Asia. Journal of Transport Geography

19:1272-1281

Lam WHK, Ng PLP, Seabrooke W, Hui CM (2004) Forecasts and reliability analysis of

port cargo throughput in Hong Kong. Urban Planning & Development Journal 130

(3):133-151

Lawrence SA (1972) International sea transport: the years ahead. Lexington Books,

Lexington/USA

Lei L, Fan C, Boile M, Theofanis S (2008) Collaborative vs. non-collaborative

container-vessel scheduling. Transportation Research Part E 44:504-520

Leung SCH, Wu Y (2004) A robust optimization model for dynamic empty container

allocation problems in an uncertain environment. International Journal of Operations

and Quantitative Management 10 (4):1-20

166

Levinson M (2008) The box: How the shipping container made the world smaller and

the world economy bigger. Princeton University Press, Princeton, New Jersey

Li X, Xu S (2011) A study on port container throughput prediction based on optimal

combined forecasting model in Shanghai port. Paper presented at the 11th

International Conference of Chinese Transportation Professionals, Nanjing, China,

Liao C-H, Tseng P-H, Cullinane K, Lu C-S (2010) The impact of an emerging port on

the carbon dioxide emissions of inland container transport: An empirical study of

Taipei port. Energy Policy 38 (9):5251-5257

Lin S (1965) Computer solutions of the traveling salesman problem. Bell Systems

Technical Journal 44:2245-2269

Lirn T, Thanopoulou H, Beynon M, Beresford A (2004) An application of AHP on

transshipment port selection: a global perspective. Maritime Economics and

Logistics 6:70-91

Liu Z, Ji L, Ye Y, Geng Z (2007) Combined forecast method of port container

throughput based on RBF neural network. Journal of Tongji University (Nature

Science) 35 (6):739-744

Løfstedt B, Alvarez JF, Plum CEM, Pisinger D, Sigurd MM (2010) An integer

programming model and benchmark suite for liner shipping network design. vol

Report 19.2010. DTU Management Engineering,

Long Y, Lee LH, Chew EP (2012) The sample average approximation method for

empty container repositioning with uncertainties. European Journal of Operational

Research 222 (1):65-75

Lu H-A, Chu C-W, Che P-Y (2010) Seasonal slot allocation planning for a container

liner shipping service. Journal of Marine Science and Technology 18 (1):84-92

Lu K, Meng Q (2011) Containership routing problem with hub-and-spoke operations.

Paper presented at the Transportation Research Board 90th Annual Meeting,

Washington/USA,

Lun YHV, Lai K-H, Cheng TCE (2010) Shipping and Logistics Management Springer-

Verlag, London

167

Luo M, Grigalunas TA (2003) A spatial-economic multimodal transportation simulation

model for US coastal container ports. Maritime Economics & Logistics 5:158-178

Mak KL, Yang DH (2007) Forecasting Hong Kong’s container throughput with

approximate least squares support vector. Paper presented at the Proceedings of the

World Congress on Engineering, London, U.K.,

McKinsey C (1967) Containerization: The Key to Low-Cost Transportation. Report to

the British Transport Docks Board,

Meng Q, Wang S (2011a) Liner shipping service network design with empty container

repositioning. Transportation Research Part E 47:695-708

Meng Q, Wang S (2012) Liner ship fleet deployment with week-dependent container

shipment demand. European Journal of Operational Research 222 (2):241-252

Meng Q, Wang S, Andersson H, Thun K (2013) Containership routing and scheduling

in liner shipping: Overview and future research directions. Transportation

Science:1–16. doi:10.1287/trsc.2013.0461

Meng Q, Wang T (2010) A chance constrained programming model for short-term liner

ship fleet planning problems. Maritime Policy & Management 37 (4):329-346

Meng Q, Wang T (2011b) A scenario-based dynamic programming model for multi-

period liner ship ﬂeet planning. Transportation Research Part E 47:401-413

Meng Q, Wang T, Gelareh S (2011) A linearized approach for the liner containership

fleet planning with demand uncertainty. In: Chew EP, Lee LH, Tang LC (eds)

Advances in Maritime Logistics and Supply Chain Systems. World Scientific

Publishing, Singapore, pp 229-256

Meng Q, Wang T, Wang S (2012) Short-term liner ship fleet planning with container

transshipment and uncertain container shipment demand. European Journal of

Operational Research 223 (1):96-105

Min H (1989) The multiple vehicle routing problem with simultaneous delivery and

pick-up points. Transportation Research Part A 23:377-386

168

Mingyong L, Erbao C (2010) An improved differential evolution algorithm for vehicle

routing problem with simultaneous pickups and deliveries and time windows.

Engineering Applications of Artificial Intelligence 23:188-195

Mladenović N, Hansen P (1997) Variable neighborhood search. Computers &

Operations Research 24:1097-1100

Montane FAT, Galvao RD (2006) A tabu search algorithm for the vehicle routing

problem with simultaneous pick-up and delivery service. European Journal of

Operational Research 33 (3):595-619

Mostafa MM (2004) Forecasting the Suez Canal traffic: a neural network analysis.

Maritime Policy & Management 31 (2):139-156

Mourao MC, Pato MV, Paixao AC (2001) Ship assignment with hub and spoke

constraints. Maritime Policy & Management 29 (2):135-150

Mulder J (2011) Constructing service networks in liner shipping. Master Thesis,

Erasmus University, Rotterdam, the Netherlands

Nagy G, Salhi S (2005) Heuristic algorithms for single and multiple depot vehicle

routing problems with pickups and deliveries. European Journal of Operational

Research 162 (126-141)

Ng AKY, Kee JKY (2008) The optimal ship sizes of container liner feeder services in

Southeast Asia: A ship operator's perspective. Maritime Policy & Management 35

(4):353-376

Notteboom T (2006) The time factor in liner shipping services. Maritime Economics

and Logistics Engineering and Management 8 (1):19-39

Notteboom TE (2004) Container shipping and ports: An overview. Review of Network

Economics 3 (2):86-106

OceanShippingConsultants (2011) South Europe and Mediterranean container markets

to 2025. Haskoning UK Ltd, Surrey

Pallis AA, de Langen PW (2010) Seaports and the structural implications of the

economic crisis. Research in Transportation Economics 27 (1):10-18

169

Pantuso G, Fagerholt K, Hvattum LM (2013) A survey on maritime fleet size and mix

problems. European Journal of Operational Research 10.1016/j.ejor.2013.04.058:1-

Parragh S, Doerner K, Hartl R (2008) A survey on pickup and delivery problems.

Journal für Betriebswirtschaft 58 (2)

Peng W-Y, Chu C-W (2009) A comparison of univariate methods for forecasting

container throughput volumes. Mathematical and Computer Modelling 50 (7-

8):1045-1057

Polat O, Günther H-O, Kulak O The containership feeder network design problem: the

new Izmir port as hub in the black sea. In: The 2th International Conference on

Logistics and Maritime Systems, Bremen, Germany, 22-24 August 2012. pp 347-

356

Polat O, Günther H-O, Kulak O (2013) The Liner Shipping Feeder Network Design

Problem. Flexible Services and Manufacturing Journal (Submitted)

Polat O, Kulak O, Günther H-O (2011) A Monte Carlo simulation based stochastic

forecasting frame: a case study of containerized trade of Turkish ports. Paper

presented at the 12th International Symposium on Econometrics Operations

Research and Statistics, Denizli, Turkey,

Polat O, Uslu EE (2010) Seasonality in Foreign Trade Data of Turkey. Gaziantep

University Journal of Social Sciences 9 (2):407-423

Qi X, Song D-P (2012) Minimizing fuel emissions by optimizing vessel schedules in

liner shipping with uncertain port times. Transportation Research Part E: Logistics

and Transportation Review 48 (4):863-880

Ran W, Xu Z, Weihong Z (2008) Analysis on comprehensive strength of Chinese

coastal container shipping company based on genetic fuzzy clustering. Paper

presented at the Proceedings of the IEEE International Conference on Automation

and Logistics, Qingdao, China September 2008

Rana K, Vickson RG (1988) A model and solution algorithm for optimal routing of a

time-chartered containership. Transportation Science 22:83-95

170

Rana K, Vickson RG (1991) Routing container ships using Lagrangean relaxation and

decomposition. Transportation Science 25:201-214

Reinhardt LB, Pisinger D (2012) A branch and cut algorithm for the container shipping

network design problem. Flexible Services and Manufacturing Journal 24 (2):349-

374

Ridolﬁ G (1999) Containerisation in the Mediterranean: Between global ocean

routeways and feeder services. GeoJournal 48:29-34

Robinson R (1998) Asian hub/feeder nets: the dynamics of restructuring. Maritime

Policy & Management 25 (1):21-40

Ronen D (1983) Cargo ship routing and scheduling: Survey of models and problems.

European Journal of Operational Research 12 (6):119-112

Ronen D (1993) Ship scheduling: The last decade. European Journal of Operational

Research 71 (3):325-333

Ronen D (2011) The effect of oil price on containership speed and fleet size. Journal of

the Operational Research Society 62:211-216

Ropke S, Pisinger D (2006) A uniﬁed heuristic for vehicle routing problems with

backhauls. European Journal of Operational Research 171:750-775

Rudić D, Hlača B (2005) Feeder service to promote revitalization of the container

transport in the Adriatic ports. Naše more, Journal of marine science and technology

52 (1-2):37-44

Rumelhart DE, Hinton GE, Williams RJ (1986) Learning internal representations by

error propagation, Vol. 1. In: Parallel distributed processing: explorations in the

microstructure of cognition. MIT Press, Cambridge (MA),

Salhi S, Nagy G (1999) A cluster insertion heuristic for single and multiple depot

vehicle routing problems with backhauling. Journal of the Operational Research

Society 50:1034-1042

171

Sambracos E, Paravantis JA, Tarantilis CD, Kiranoudis CT (2004) Dispatching of small

containers via coastal freight liners: The case of the Aegean Sea. European Journal

of Operational Research 152:365-381

Schulze PM, Prinz A (2009) Forecasting container transshipment in Germany. Applied

Economics 41:2809-2815

Seabrooke W, Hui ECM, Lam WHK, Wong GKC (2003) Forecasting cargo growth and

regional role of the port of Hong Kong. Cities 20 (1):51-64

Shintani K, Imai A, Nishimura E, Papadimitriou S (2007) The container shipping

network design problem with empty container repositioning. Transportation

Research Part E 43 (1):39-59

ShippingChina (2013) China Forwarders Freight Index (CFFI).

http://en.shippingchina.com/scfi/index/index.html. Accessed 19.03.2013

Sigurt M, Ulstein N, Nygreen B, Ryan D (2005) Ship scheduling with recurring visits

and visit separation requirements. In: Desaulniers G, Desrosiers, J., Solomon, M.

(ed) Column Generation. Springer, US, pp 225-245

Song D-P, Dong J-X (2008) Empty container management in cyclic shipping routes.

Maritime Economics & Logistics 10:335–361

Stenger A, Vigo D, Enz S, Schwind M (2012) An adaptive variable neighborhood

search algorithm for a vehicle routing problem arising in small package shipping.

Transportation Science doi:10.1287/trsc.1110.0396

Stopford M (2009) Maritime economics. Third Edition edn. Routledge, New York

Styhre L (2010) Capacity utilisation in short sea shipping (Chapter 9). PhD Thesis,

Chalmers University of Technology, Göteborg, Sweden

Subramanian A, Cabral LdAF (2008) An ILS Based Heuristic for the Vehicle Routing

Problem with Simultaneous Pickup and Delivery and Time Limit. Lecture Notes in

Computer Science 4972:135-146

172

Subramanian A, Drummond LMA, Bentes C, Ochi LS, Farias R (2010) A parallel

heuristic for the vehicle routing problem with simultaneous pickup and delivery.

Computers and Operations Research 37:1899-1911

Sun L (2010) The research on a double forecasting model of port cargo throughput.

World Journal of Modeling and Simulation 6 (1):57-62

Sun X, Li N (2006) A new immune genetic algorithm for the large scale ship routing

problems. Paper presented at the International Conference on Management of

Logistics and Supply Chain Chang Sha/Chine - Sydney/Australia,

Sys C, Blauwens G, Omey E, Voorde EVD, Witlox F (2008) In Search of the Link

between Ship Size and Operations. Transportation Planning and Technology 31

(4):435-463

Takano K, Arai M (2009) A genetic algorithm for the hub-and-spoke problem applied

to containerized cargo transport. Journal of Marine Science and Technology 14:256-

274

Tang FA, Galvao RD (2006) A tabu search algorithm for the vehicle routing problem

with simultaneous pick-up and delivery service. Computer and Operations Research

33:595-619

Ting S-C, Tzeng G-H (2003) Ship scheduling and cost analysis for route planning in

liner shipping. Maritime Economics & Logistics 5 (4):378-392

UNCTAD (2012) United Nations conference on trade and development review of

maritime transport. United Nations, Newyork and Geneva

Varbanova A (2011a) Current iIssues in operational planning of general cargo

transportation on container feeder lines in the Black Sea region. The international

virtual journal for science, techniques and innovations for the industry "Machines,

Technologies, Materials” 2011 (3):35-38

Varbanova A (2011b) Theoretical analysis of cost efficiency of general cargo

transportation on container feeder lines in the Black Sea region. The international

virtual journal for science, techniques and innovations for the industry "Machines,

Technologies, Materials” 2011 (3):51-53

173

VHSS (2013) Containership time-charter-rates. http://www.vhss.de/contex_new.php.

Accessed 15.06.2012

Walter M, Younger W (1988) Forecasting the demand for services of a new port.

GeoJournal 16 (3):295-300

Wang B, Tang G (2010) Stochastic Optimization Model for Container Shipping of Sea

Carriage. Journal of Transportation Systems Engineering and Information

Technology 10 (3):58-63

Wang S, Meng Q (2011) Liner shipping network design with time window constraints.

Paper presented at the Transportation Research Board 90th Annual Meeting,

Washington/USA,

Wang S, Meng Q (2012a) Liner ship fleet deployment with container transshipment

operations. Transportation Research Part E 48 (2):470-484

Wang S, Meng Q (2012b) Liner ship route schedule design with sea contingency time

and port time uncertainty. Transportation Research Part B: Methodological 46

(5):615-633

Wang S, Meng Q (2012c) Robust schedule design for liner shipping services.

Transportation Research Part E: Logistics and Transportation Review 48 (6):1093-

1106

Wang S, Meng Q, Bell MGH (2013a) Liner ship route capacity utilization estimation

with a bounded polyhedral container shipment demand pattern. Transportation

Research Part B: Methodological 47:57-76

Wang S, Meng Q, Liu Z (2013b) Bunker consumption optimization methods in

shipping: A critical review and extensions. Transportation Research Part E:

Logistics and Transportation Review 53:49-62

Wang S, Meng Q, Sun Z (2013c) Container routing in liner shipping. Transportation

Research Part E: Logistics and Transportation Review 49 (1):1-7

Wang S, Wang T, Meng Q (2011) A note on liner ship fleet deployment. Flexible

Services and Manufacturing Journal 23 (4):422-430

174

Wang T, Meng Q, Wang S (2012) Robust Optimization Model for Liner Ship Fleet

Planning with Container Transshipment and Uncertain Demand. Transportation

Research Record 2273:18-28

Wassan NA, Wassan AH, Nagy G (2008) A reactive tabu search algorithm for the

vehicle routing problem with simultaneous pickups and deliveries. Journal of

Combinatorial Optimization 15:368-386

Waters W (1973) Coastal Container Shipping. The Journal of the New Zealand

Instituaon of Engineering 28 (10):281-287

Wijnolst N, Waals F, Bello F, Gendronneau Y, Kempen Dv (2000) Malacca Max (2)

Container Shipping Network Economy. Delft University Press, Delft/Netherlands

Wikipedia (2013a) Container ship. http://en.wikipedia.org/wiki/Container_ship.

Accessed 11.03.2013

Wikipedia (2013b) Intermodal container.

http://en.wikipedia.org/wiki/Intermodal_container. Accessed 11.03.2013

Wikipedia (2013c) Liner service. http://en.wikipedia.org/wiki/Liner_service. Accessed

14.03.2013

Wilmsmeier G, Notteboom T (2011) Determinants of liner shippping network

conﬁguration: A two-region comparison. GeoJournal 76 (3):213-228

Windeck V (2013) A liner shipping network design: Routing and scheduling

considering environmental influences. Produktion und Logistik. Springer Gabler,

Wiesbaden

WorldShipping (2013a) Benefits of liner shipping: Efficiency.

http://www.worldshipping.org/benefits-of-liner-shipping/efficiency. Accessed

15.03.2013

WorldShipping (2013b) Benefits of liner shipping: Low environmental impact.

http://www.worldshipping.org/benefits-of-liner-shipping/low-environmental-impact.

Accessed 11.03.2013

175

WorldShipping (2013c) Global trade: Landside connections.

http://www.worldshipping.org/about-the-industry/global-trade/inland-dispersal-of-

cargo. Accessed 14.03.2013

Wu D, Pan X (2010) Container volume forecasting of Jiujiang port based on SVM and

game theory. Paper presented at the International Conference on Intelligent

Computation Technology and Automation, Hunan, China.,

Xiao Y, Xiao J, Wang S (2012) A hybrid forecasting model for non-stationary time

series: An application to container throughput prediction. International Journal of

Knowledge and Systems Science 3 (2):1-16

Yan S, Chen C-Y, Lin S-C (2009) Ship scheduling and container shipment planning for

liners in short-term operations. Journal of Marine Science and Technology 14

(4):417-435

Yang Z, Chen K (2010) Optimization of shipping network of trunk and feeder lines for

inter-regional and intra-regional container transport Journal of the Eastern Asia

Society for Transportation Studies 8:682-693

Yang Z, Chen K, Notteboom T (2012) Optimal design of container liner services:

Interactions with the transport demand in ports. Maritime Economics & Logistics 14

(4):409-434

Zachariadis EE, Kiranoudis CT (2011) A local search metaheuristic algorithm for the

vehiclerouting problem with simultaneous pick-ups and deliveries. Expert Systems

with Applications 38 (3):2717-2726

Zachariadis EE, Tarantilis CD, Kiranoudis CT (2009) A hybrid metaheuristic algorithm

for the vehiclerouting problem with simultaneous delivery and pick-up service.

Expert Systems with Applications 36 ( 2/1):1070-1081

Zacharioudakis PG, Iordanis S, Lyridis DV, Psaraftis HN (2011) Liner shipping cycle

cost modelling, fleet deployment optimization and what-if analysis. Maritime

Economics & Logistics 13:278-297

Zachcial M, Lemper B (2006) Container shipping: An overview of Development

Trends. In: Heideloff C, Pawlik T (eds) Handbook of Container Shipping

176

Management. vol 32. Institute of Shipping Economics and Logistics (ISL), Bremen,

Germany, pp 23-37

Zeng Q, Yang Z, Chen C (2010) Robust optimization model for resource allocation of

container shipping lines. Tsinghua Science & Technology 15 (5):586-594

Zeng Z, Yang Z (2002) Dynamic programming of port position and scale in the

hierarchized container ports network. Maritime Policy & Management 29 (2):163-

177

Zhang G, Patuwo BE, Hu MY (1998) Forecasting with artificial neural networks: The

state of the art. International Journal of Forecasting 14 (1):35-62

Zhang P, Cui Y (2011) Research on combination forecast of port container throughput

based on Elman neural network. Paper presented at the 3rd International Conference

on Communication Software and Networks, Xi'an, China,

Zhang T, Tian W-x, Zhang Y-j, Liu S-x (2008) Improved ant colony system for

VRPSPD with maximum distance constraint. Systems Engineering - Theory &

Practice 28 (1):132-140

Zohil J, Prijon M (1999) The MED rule: the interdependence of container throughput

and transhipment volumes in the Mediterranean ports. Maritime Policy &

Management 26 (2):175-193

Zurada JM (1992) An Introduction to artificial Neural systems. West Publishing, St.

Paul

Zurheide S, Fischer K (2011) A Simulation study for evaluating a slot allocation model

for a liner shipping network. Lecture Notes in Computer Science 6971:354-369

Zurheide S, Fischer K (2012) A revenue management slot allocation model for liner

shipping networks. Maritime Economics & Logistics 14:334-361