ScienceDirect
Available online at www.sciencedirect.com
Procedia Computer Science 151 (2019) 820–825
1877-0509 © 2019 The Authors. Published by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Peer-review under responsibility of the Conference Program Chairs.
10.1016/j.procs.2019.04.112
10.1016/j.procs.2019.04.112 1877-0509
Available online at www.sciencedirect.com
Procedia Computer Science 00 (2019) 000–000
www.elsevier.com/locate/procedia
The 8th International Workshop on Agent-based Mobility, Traffic and Transportation Models,
Methodologies and Applications (ABMTRANS 2019),
April 29 - May 2, 2019, Leuven, Belgium
Large-Scale Assignment of Congested Bicycle Traffic Using Speed
Heterogeneous Agents
Mads Paulsena,∗, Kai Nagelb
aTechnical University of Denmark, Department of Management Engineering, Bygningstorvet 116B, 2800 Kgs. Lyngby, Denmark
bTechnische Universit¨at Berlin, Transport Systems Planning and Telematics, Salzufer 17-19, 10587 Berlin, Germany
Abstract
Despite requiring less space than most other modes of transport, bicycle traffic will also be prone to congestion when the traffic
volume is sufficiently large. Such congestion can eventually influence the route choices of cyclists using the network. In this study
we model bicycle congestion on a detailed network of the greater Copenhagen area by assigning an entire day of bicycle traffic
using a recently developed method for dynamic network loading of speed heterogeneous multi-lane bicycle traffic. The model
iteratively assigns appropriate routes for more than a million bicycle trips in the demand sensitive network, and with computation
times of less than 15 minutes per iteration the proposed model proves to be large-scale applicable. This makes it the first dedicated
bicycle traffic assignment model to account for congestion. The results indicate that the solid bicycle infrastructure of Copenhagen
and cyclists’ willingness to change routes are key to keeping travel times low for cyclists.
c
2019 The Authors. Published by Elsevier B.V.
Peer-review under responsibility of the Conference Program Chairs.
Keywords: Bicycle traffic assignment, bicycle congestion modelling, multi-agent simulation, speed heterogeneity
1. Introduction
Although congestion on bicycle tracks may seem unlikely many places in the world, in some cities (e.g. Copen-
hagen) the traffic level of bicycles has reached a level which leads to considerable bicycle congestion. The congestion
influences the travel time, and may also influence cyclists’ route choice.
For decades the literature1has been concerned with how the level-of-service for cyclists is influenced by the car
traffic volume. However, including the flow of bicycles as a parameter of attractiveness of a route has only received
little attention in the literature. Two stated preferences studies2,3 found that cyclists generally have disutility towards
increasing bicycle flow on bicycle lanes in Nanjing, China. This indicates that the more cyclists using a link, the
less attractive the link will become. In traffic assignment for motorised traffic such disutility can be explained by
∗Corresponding author. Tel.: +45 45 25 65 98.
1877-0509 c
2019 The Authors. Published by Elsevier B.V.
Peer-review under responsibility of the Conference Program Chairs.
Available online at www.sciencedirect.com
Procedia Computer Science 00 (2019) 000–000
www.elsevier.com/locate/procedia
The 8th International Workshop on Agent-based Mobility, Traffic and Transportation Models,
Methodologies and Applications (ABMTRANS 2019),
April 29 - May 2, 2019, Leuven, Belgium
Large-Scale Assignment of Congested Bicycle Traffic Using Speed
Heterogeneous Agents
Mads Paulsena,∗, Kai Nagelb
aTechnical University of Denmark, Department of Management Engineering, Bygningstorvet 116B, 2800 Kgs. Lyngby, Denmark
bTechnische Universit¨at Berlin, Transport Systems Planning and Telematics, Salzufer 17-19, 10587 Berlin, Germany
Abstract
Despite requiring less space than most other modes of transport, bicycle traffic will also be prone to congestion when the traffic
volume is sufficiently large. Such congestion can eventually influence the route choices of cyclists using the network. In this study
we model bicycle congestion on a detailed network of the greater Copenhagen area by assigning an entire day of bicycle traffic
using a recently developed method for dynamic network loading of speed heterogeneous multi-lane bicycle traffic. The model
iteratively assigns appropriate routes for more than a million bicycle trips in the demand sensitive network, and with computation
times of less than 15 minutes per iteration the proposed model proves to be large-scale applicable. This makes it the first dedicated
bicycle traffic assignment model to account for congestion. The results indicate that the solid bicycle infrastructure of Copenhagen
and cyclists’ willingness to change routes are key to keeping travel times low for cyclists.
c
2019 The Authors. Published by Elsevier B.V.
Peer-review under responsibility of the Conference Program Chairs.
Keywords: Bicycle traffic assignment, bicycle congestion modelling, multi-agent simulation, speed heterogeneity
1. Introduction
Although congestion on bicycle tracks may seem unlikely many places in the world, in some cities (e.g. Copen-
hagen) the traffic level of bicycles has reached a level which leads to considerable bicycle congestion. The congestion
influences the travel time, and may also influence cyclists’ route choice.
For decades the literature1has been concerned with how the level-of-service for cyclists is influenced by the car
traffic volume. However, including the flow of bicycles as a parameter of attractiveness of a route has only received
little attention in the literature. Two stated preferences studies2,3 found that cyclists generally have disutility towards
increasing bicycle flow on bicycle lanes in Nanjing, China. This indicates that the more cyclists using a link, the
less attractive the link will become. In traffic assignment for motorised traffic such disutility can be explained by
∗Corresponding author. Tel.: +45 45 25 65 98.
1877-0509 c
2019 The Authors. Published by Elsevier B.V.
Peer-review under responsibility of the Conference Program Chairs.
Available online at www.sciencedirect.com
Procedia Computer Science 00 (2019) 000–000
www.elsevier.com/locate/procedia
The 8th International Workshop on Agent-based Mobility, Traffic and Transportation Models,
Methodologies and Applications (ABMTRANS 2019),
April 29 - May 2, 2019, Leuven, Belgium
Large-Scale Assignment of Congested Bicycle Traffic Using Speed
Heterogeneous Agents
Mads Paulsena,∗, Kai Nagelb
aTechnical University of Denmark, Department of Management Engineering, Bygningstorvet 116B, 2800 Kgs. Lyngby, Denmark
bTechnische Universit¨at Berlin, Transport Systems Planning and Telematics, Salzufer 17-19, 10587 Berlin, Germany
Abstract
Despite requiring less space than most other modes of transport, bicycle traffic will also be prone to congestion when the traffic
volume is sufficiently large. Such congestion can eventually influence the route choices of cyclists using the network. In this study
we model bicycle congestion on a detailed network of the greater Copenhagen area by assigning an entire day of bicycle traffic
using a recently developed method for dynamic network loading of speed heterogeneous multi-lane bicycle traffic. The model
iteratively assigns appropriate routes for more than a million bicycle trips in the demand sensitive network, and with computation
times of less than 15 minutes per iteration the proposed model proves to be large-scale applicable. This makes it the first dedicated
bicycle traffic assignment model to account for congestion. The results indicate that the solid bicycle infrastructure of Copenhagen
and cyclists’ willingness to change routes are key to keeping travel times low for cyclists.
c
2019 The Authors. Published by Elsevier B.V.
Peer-review under responsibility of the Conference Program Chairs.
Keywords: Bicycle traffic assignment, bicycle congestion modelling, multi-agent simulation, speed heterogeneity
1. Introduction
Although congestion on bicycle tracks may seem unlikely many places in the world, in some cities (e.g. Copen-
hagen) the traffic level of bicycles has reached a level which leads to considerable bicycle congestion. The congestion
influences the travel time, and may also influence cyclists’ route choice.
For decades the literature1has been concerned with how the level-of-service for cyclists is influenced by the car
traffic volume. However, including the flow of bicycles as a parameter of attractiveness of a route has only received
little attention in the literature. Two stated preferences studies2,3 found that cyclists generally have disutility towards
increasing bicycle flow on bicycle lanes in Nanjing, China. This indicates that the more cyclists using a link, the
less attractive the link will become. In traffic assignment for motorised traffic such disutility can be explained by
∗Corresponding author. Tel.: +45 45 25 65 98.
1877-0509 c
2019 The Authors. Published by Elsevier B.V.
Peer-review under responsibility of the Conference Program Chairs.
2M. Paulsen et al. /Procedia Computer Science 00 (2019) 000–000
increased travel time modelled with a flow-dependent travel time function for each link4. However, for bicycle traffic
assignment only a single model5has so far incorporated a mechanism that makes links with high flows less attractive.
Primarily dealing with determining the static link characteristics influencing bicycle route choice, a wide range of
dedicated bicycle route choice models exist6– some of which also perform one-shot assignments of bicycle flows7,8.
Other studies9,10,11,12,13 have focused on assigning bicycles traffic onto networks including various link characteristics,
but without any feedback from the network such as lowered safety or speeds influencing the route choice, although
one of the studies12 mentions it as a necessary extension. Another study14 solely dealing with route choice also argues
– as a suggestion for future research – that bicycle route choice models ought to incorporate congestion effects from
traffic assignment models with dedicated bicycle volume-delay functions.
Nevertheless, until now the literature only contains one example5of such bicycle traffic assignment considering
congestion effects. That study focuses on mixed traffic in Patna, India and assigns mixed traffic onto a multi-modal,
shared network whilst taking the maximum speeds of trucks, cars, motorbikes, and bicycles into account. However,
due to the chaotic Indian traffic scheme with mixed traffic being far from what is seen in a city with plenty of dedicated
bicycle infrastructure, there still is a need for a dedicated bicycle traffic assignment model that can model congestion
in segregated bicycle traffic.
The aim of this study is to fill this gap by assigning bicycle traffic onto a network at a large scale using an iterative
feedback loop between the supply and the demand, as it is commonly known from motorised traffic. Based on a
recently developed network loading model for heterogeneous bicycle traffic15 this study includes detailed congestion
patterns on link level, and forms a complete agent-based traffic assignment model for bicycles by allowing every
cyclist to adjust his/her route according to the congested network.
The study does not deal with travel time delays caused by conflicting traffic at intersections nor traffic signals.
We acknowledge that such delays cause substantial added travel time for cyclists. However, in order to keep focus
on the large-scale applicability and consequences of the implemented network loading model such delays have been
considered to be beyond the scope of this paper.
The remainder of the paper is structured as follows. Section 2 contains the applied methodology including a brief
summary of the used network loading model15. The data used for the case study is described in Section 3, before
presenting the results of said case study in Section 4. Finally, a discussion of the realism and the limitations of the
model is found in Section 5 alongside suggestions for future research.
2. Methodology
In order to have an assignment model for bicycle traffic, it is necessary to be able to model how the demand
influences the supply and vice versa. Plenty of route choice models for bicycle traffic exist6, meaning that it is known
to a large extend how the supply influences the demand for any link in the network. Until recently, however, it has
been ignored how the supply is influenced by the demand predicted by such route choice models.
The network loading used in the assignment model of this study is a MATSim16-implementation of a recently
proposed methodology15 for dynamic, large-scale applicable network loading of bicycle traffic. The model is based on
individual desired speeds for every cyclist corresponding to an individualised free-flow speed. It supports multi-lane
links with each link having a predefined number of pseudo-lanes, allowing faster cyclists to overtake slower cyclists
by placing themselves in the outermost lane(s) as long as there is room. If a cyclist can no longer choose a pseudo-
lane that will satisfy his/her desired speed, the cyclist will choose the fastest available pseudo-lane. Furthermore, each
cyclist has an individualised, speed dependent headway distance, that he/she must keep to the cyclist in front.
The speed of a cyclist entering a lane can be determined at link entrance solely based on his/her desired speed
and headway preferences, as well as the time at which the previous entrant of the selected pseudo-lane is going to be
leaving and entered the link, as neither overtaking nor lane changing are allowed within a lane. Because delays are
based on the lack of opportunity to overtake, it also means that cyclists with high desired speeds have a higher tendency
to be delayed, whereas it is virtually impossible for the overall slowest cyclist to be delayed. The methodology yields
reasonable aggregate fundamental diagrams15 as long as there is sufficient speed heterogeneity, and as such is best
suited for agent-based transport modelling where individualised preferences can be assigned directly to each agent.
The route choice model used in this study is a simple multinomial logit model where the free flow travel time and
congested travel time of alternatives are the only two variables used in the disutility function. The congested travel
© 2019 The Authors. Published by Elsevier B.V.
This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Peer-review under responsibility of the Conference Program Chairs.
Mads Paulsen et al. / Procedia Computer Science 151 (2019) 820–825 821
Available online at www.sciencedirect.com
Procedia Computer Science 00 (2019) 000–000
www.elsevier.com/locate/procedia
The 8th International Workshop on Agent-based Mobility, Traffic and Transportation Models,
Methodologies and Applications (ABMTRANS 2019),
April 29 - May 2, 2019, Leuven, Belgium
Large-Scale Assignment of Congested Bicycle Traffic Using Speed
Heterogeneous Agents
Mads Paulsena,∗, Kai Nagelb
aTechnical University of Denmark, Department of Management Engineering, Bygningstorvet 116B, 2800 Kgs. Lyngby, Denmark
bTechnische Universit¨at Berlin, Transport Systems Planning and Telematics, Salzufer 17-19, 10587 Berlin, Germany
Abstract
Despite requiring less space than most other modes of transport, bicycle traffic will also be prone to congestion when the traffic
volume is sufficiently large. Such congestion can eventually influence the route choices of cyclists using the network. In this study
we model bicycle congestion on a detailed network of the greater Copenhagen area by assigning an entire day of bicycle traffic
using a recently developed method for dynamic network loading of speed heterogeneous multi-lane bicycle traffic. The model
iteratively assigns appropriate routes for more than a million bicycle trips in the demand sensitive network, and with computation
times of less than 15 minutes per iteration the proposed model proves to be large-scale applicable. This makes it the first dedicated
bicycle traffic assignment model to account for congestion. The results indicate that the solid bicycle infrastructure of Copenhagen
and cyclists’ willingness to change routes are key to keeping travel times low for cyclists.
c
2019 The Authors. Published by Elsevier B.V.
Peer-review under responsibility of the Conference Program Chairs.
Keywords: Bicycle traffic assignment, bicycle congestion modelling, multi-agent simulation, speed heterogeneity
1. Introduction
Although congestion on bicycle tracks may seem unlikely many places in the world, in some cities (e.g. Copen-
hagen) the traffic level of bicycles has reached a level which leads to considerable bicycle congestion. The congestion
influences the travel time, and may also influence cyclists’ route choice.
For decades the literature1has been concerned with how the level-of-service for cyclists is influenced by the car
traffic volume. However, including the flow of bicycles as a parameter of attractiveness of a route has only received
little attention in the literature. Two stated preferences studies2,3 found that cyclists generally have disutility towards
increasing bicycle flow on bicycle lanes in Nanjing, China. This indicates that the more cyclists using a link, the
less attractive the link will become. In traffic assignment for motorised traffic such disutility can be explained by
∗Corresponding author. Tel.: +45 45 25 65 98.
1877-0509 c
2019 The Authors. Published by Elsevier B.V.
Peer-review under responsibility of the Conference Program Chairs.
Available online at www.sciencedirect.com
Procedia Computer Science 00 (2019) 000–000
www.elsevier.com/locate/procedia
The 8th International Workshop on Agent-based Mobility, Traffic and Transportation Models,
Methodologies and Applications (ABMTRANS 2019),
April 29 - May 2, 2019, Leuven, Belgium
Large-Scale Assignment of Congested Bicycle Traffic Using Speed
Heterogeneous Agents
Mads Paulsena,∗, Kai Nagelb
aTechnical University of Denmark, Department of Management Engineering, Bygningstorvet 116B, 2800 Kgs. Lyngby, Denmark
bTechnische Universit¨at Berlin, Transport Systems Planning and Telematics, Salzufer 17-19, 10587 Berlin, Germany
Abstract
Despite requiring less space than most other modes of transport, bicycle traffic will also be prone to congestion when the traffic
volume is sufficiently large. Such congestion can eventually influence the route choices of cyclists using the network. In this study
we model bicycle congestion on a detailed network of the greater Copenhagen area by assigning an entire day of bicycle traffic
using a recently developed method for dynamic network loading of speed heterogeneous multi-lane bicycle traffic. The model
iteratively assigns appropriate routes for more than a million bicycle trips in the demand sensitive network, and with computation
times of less than 15 minutes per iteration the proposed model proves to be large-scale applicable. This makes it the first dedicated
bicycle traffic assignment model to account for congestion. The results indicate that the solid bicycle infrastructure of Copenhagen
and cyclists’ willingness to change routes are key to keeping travel times low for cyclists.
c
2019 The Authors. Published by Elsevier B.V.
Peer-review under responsibility of the Conference Program Chairs.
Keywords: Bicycle traffic assignment, bicycle congestion modelling, multi-agent simulation, speed heterogeneity
1. Introduction
Although congestion on bicycle tracks may seem unlikely many places in the world, in some cities (e.g. Copen-
hagen) the traffic level of bicycles has reached a level which leads to considerable bicycle congestion. The congestion
influences the travel time, and may also influence cyclists’ route choice.
For decades the literature1has been concerned with how the level-of-service for cyclists is influenced by the car
traffic volume. However, including the flow of bicycles as a parameter of attractiveness of a route has only received
little attention in the literature. Two stated preferences studies2,3 found that cyclists generally have disutility towards
increasing bicycle flow on bicycle lanes in Nanjing, China. This indicates that the more cyclists using a link, the
less attractive the link will become. In traffic assignment for motorised traffic such disutility can be explained by
∗Corresponding author. Tel.: +45 45 25 65 98.
1877-0509 c
2019 The Authors. Published by Elsevier B.V.
Peer-review under responsibility of the Conference Program Chairs.
Available online at www.sciencedirect.com
Procedia Computer Science 00 (2019) 000–000
www.elsevier.com/locate/procedia
The 8th International Workshop on Agent-based Mobility, Traffic and Transportation Models,
Methodologies and Applications (ABMTRANS 2019),
April 29 - May 2, 2019, Leuven, Belgium
Large-Scale Assignment of Congested Bicycle Traffic Using Speed
Heterogeneous Agents
Mads Paulsena,∗, Kai Nagelb
aTechnical University of Denmark, Department of Management Engineering, Bygningstorvet 116B, 2800 Kgs. Lyngby, Denmark
bTechnische Universit¨at Berlin, Transport Systems Planning and Telematics, Salzufer 17-19, 10587 Berlin, Germany
Abstract
Despite requiring less space than most other modes of transport, bicycle traffic will also be prone to congestion when the traffic
volume is sufficiently large. Such congestion can eventually influence the route choices of cyclists using the network. In this study
we model bicycle congestion on a detailed network of the greater Copenhagen area by assigning an entire day of bicycle traffic
using a recently developed method for dynamic network loading of speed heterogeneous multi-lane bicycle traffic. The model
iteratively assigns appropriate routes for more than a million bicycle trips in the demand sensitive network, and with computation
times of less than 15 minutes per iteration the proposed model proves to be large-scale applicable. This makes it the first dedicated
bicycle traffic assignment model to account for congestion. The results indicate that the solid bicycle infrastructure of Copenhagen
and cyclists’ willingness to change routes are key to keeping travel times low for cyclists.
c
2019 The Authors. Published by Elsevier B.V.
Peer-review under responsibility of the Conference Program Chairs.
Keywords: Bicycle traffic assignment, bicycle congestion modelling, multi-agent simulation, speed heterogeneity
1. Introduction
Although congestion on bicycle tracks may seem unlikely many places in the world, in some cities (e.g. Copen-
hagen) the traffic level of bicycles has reached a level which leads to considerable bicycle congestion. The congestion
influences the travel time, and may also influence cyclists’ route choice.
For decades the literature1has been concerned with how the level-of-service for cyclists is influenced by the car
traffic volume. However, including the flow of bicycles as a parameter of attractiveness of a route has only received
little attention in the literature. Two stated preferences studies2,3 found that cyclists generally have disutility towards
increasing bicycle flow on bicycle lanes in Nanjing, China. This indicates that the more cyclists using a link, the
less attractive the link will become. In traffic assignment for motorised traffic such disutility can be explained by
∗Corresponding author. Tel.: +45 45 25 65 98.
1877-0509 c
2019 The Authors. Published by Elsevier B.V.
Peer-review under responsibility of the Conference Program Chairs.
2M. Paulsen et al. /Procedia Computer Science 00 (2019) 000–000
increased travel time modelled with a flow-dependent travel time function for each link4. However, for bicycle traffic
assignment only a single model5has so far incorporated a mechanism that makes links with high flows less attractive.
Primarily dealing with determining the static link characteristics influencing bicycle route choice, a wide range of
dedicated bicycle route choice models exist6– some of which also perform one-shot assignments of bicycle flows7,8.
Other studies9,10,11,12,13 have focused on assigning bicycles traffic onto networks including various link characteristics,
but without any feedback from the network such as lowered safety or speeds influencing the route choice, although
one of the studies12 mentions it as a necessary extension. Another study14 solely dealing with route choice also argues
– as a suggestion for future research – that bicycle route choice models ought to incorporate congestion effects from
traffic assignment models with dedicated bicycle volume-delay functions.
Nevertheless, until now the literature only contains one example5of such bicycle traffic assignment considering
congestion effects. That study focuses on mixed traffic in Patna, India and assigns mixed traffic onto a multi-modal,
shared network whilst taking the maximum speeds of trucks, cars, motorbikes, and bicycles into account. However,
due to the chaotic Indian traffic scheme with mixed traffic being far from what is seen in a city with plenty of dedicated
bicycle infrastructure, there still is a need for a dedicated bicycle traffic assignment model that can model congestion
in segregated bicycle traffic.
The aim of this study is to fill this gap by assigning bicycle traffic onto a network at a large scale using an iterative
feedback loop between the supply and the demand, as it is commonly known from motorised traffic. Based on a
recently developed network loading model for heterogeneous bicycle traffic15 this study includes detailed congestion
patterns on link level, and forms a complete agent-based traffic assignment model for bicycles by allowing every
cyclist to adjust his/her route according to the congested network.
The study does not deal with travel time delays caused by conflicting traffic at intersections nor traffic signals.
We acknowledge that such delays cause substantial added travel time for cyclists. However, in order to keep focus
on the large-scale applicability and consequences of the implemented network loading model such delays have been
considered to be beyond the scope of this paper.
The remainder of the paper is structured as follows. Section 2 contains the applied methodology including a brief
summary of the used network loading model15. The data used for the case study is described in Section 3, before
presenting the results of said case study in Section 4. Finally, a discussion of the realism and the limitations of the
model is found in Section 5 alongside suggestions for future research.
2. Methodology
In order to have an assignment model for bicycle traffic, it is necessary to be able to model how the demand
influences the supply and vice versa. Plenty of route choice models for bicycle traffic exist6, meaning that it is known
to a large extend how the supply influences the demand for any link in the network. Until recently, however, it has
been ignored how the supply is influenced by the demand predicted by such route choice models.
The network loading used in the assignment model of this study is a MATSim16-implementation of a recently
proposed methodology15 for dynamic, large-scale applicable network loading of bicycle traffic. The model is based on
individual desired speeds for every cyclist corresponding to an individualised free-flow speed. It supports multi-lane
links with each link having a predefined number of pseudo-lanes, allowing faster cyclists to overtake slower cyclists
by placing themselves in the outermost lane(s) as long as there is room. If a cyclist can no longer choose a pseudo-
lane that will satisfy his/her desired speed, the cyclist will choose the fastest available pseudo-lane. Furthermore, each
cyclist has an individualised, speed dependent headway distance, that he/she must keep to the cyclist in front.
The speed of a cyclist entering a lane can be determined at link entrance solely based on his/her desired speed
and headway preferences, as well as the time at which the previous entrant of the selected pseudo-lane is going to be
leaving and entered the link, as neither overtaking nor lane changing are allowed within a lane. Because delays are
based on the lack of opportunity to overtake, it also means that cyclists with high desired speeds have a higher tendency
to be delayed, whereas it is virtually impossible for the overall slowest cyclist to be delayed. The methodology yields
reasonable aggregate fundamental diagrams15 as long as there is sufficient speed heterogeneity, and as such is best
suited for agent-based transport modelling where individualised preferences can be assigned directly to each agent.
The route choice model used in this study is a simple multinomial logit model where the free flow travel time and
congested travel time of alternatives are the only two variables used in the disutility function. The congested travel
822 Mads Paulsen et al. / Procedia Computer Science 151 (2019) 820–825
M. Paulsen et al. /Procedia Computer Science 00 (2019) 000–000 3
time, i.e. the difference between the actual travel time and the free flow travel time, is penalised 50 % harder than free
flow travel time, in line with previous findings for car users in the area17.
Each agent has a maximum number of five plans in his/her choice set, and after each iteration the performed plan
is assigned a score based on the obtained travel time in the previous iteration. New plans are added to the choice
set by performing a shortest path search in a network with empirical average travel times in time bins of 15 minutes
based on the previous iteration. However, during the shortest path search the link cost of every link is agent-specific
by using the maximum between the empirical (time binned) travel time and the free flow travel time based on that
agent’s desired speed.
Although the framework allows extension to include other link (to link) attributes previously seen in the literature
such as signals, slope, surface, and left/right turns, for the time being only travel time is used in order to focus on the
congestion caused by the network loading model.
The traffic assignment model has been implemented in the open-source and agent-based transport simulation soft-
ware MATSim16 by replacing parts of the default mobility simulator with the model from Paulsen et al. (2018)15,
alongside minor changes in the routing in order to account for individual desired speeds.
3. Case Study
In order to test the realism and the large-scale applicability of the model a case study for the greater Copenhagen
area is carried out. The demand is based on the Copenhagen Model for Person Activity Scheduling (COMPAS)18,
producing daily activity plans for a synthetic population of the area. Only persons with a bicycle trip during his/her
day have been used for this study, resulting in a population of 547,085 persons with daily activity plans forming a
total of 1,082,958 bicycle trips.
Based on aerial data of observed, uncongested bicycle traffic on Smallegade in Frederiksberg, every individual is
assigned a normally distributed headway distance parameter and a desired speed (v0in m/s) based on Johnson’s SU
distribution19 with the estimated parameters γ=−2.75, ξ=3.67, δ=4.07, and λ=3.49,15
v0∼δ
λ√2π1+v0−ξ
λ2e−1
2γ+δsinh−1v0−ξ
λ2
,(1)
inferring a mean speed and standard deviation of roughly 22 km/h and 4 km/h, respectively. This distribution has the
highest likelihood and lowest Kolmogorov-Snirnov statistic among 11 candidate distributions with estimated parame-
ters, and does not deviate significantly from the empirical distribution from Smallegade according to the Kolmogorov-
Smirnov Goodness-of-Fit test (p-value above 0.6).15
The network is based on OpenStreetMap (OSM)20 and is included in MATSim by altering the default OSM network
reader21. All link types that generally allow bicycle traffic have been included in the model unless the road explicitly
stated that bicycle traffic was not allowed. Furthermore, any road where bicycle traffic is explicitly mentioned as
allowed or designated has also been included in the network.
The network loading model15 only needs the lengths and widths of the bicycle infrastructure of links. Lengths
are always available and can always be extracted directly from the OSM data. Widths also play a vital part for the
methodology, but are rarely available. On the few links where widths (ωl) were available, the number of lanes were
determined based on the widths according to a Danish study22, that found the number of efficient lanes (Ψl) of bicycle
traffic to be determined by,
Ψl=1+ωl−0.4m
1.25 m.(2)
Widths were manually added to crucial arterial roads of Copenhagen – especially the cycle superhighways – before
extracting the data from OSM. This was done in order to secure these links being capable of handing large amount of
bicycle traffic. Where widths were not available, the type of bicycle infrastructure determined the number of lanes.
Explicit cycleways where motorised traffic is not allowed, as well as roads with bicycles riding in lanes or tracks were
given two pseudo-lanes, since practically all of such infrastructure in Denmark would be wide enough to form two
efficient lanes. Roads with no information about cycleways and footpaths where bicycle traffic is allowed were only
given a single lane, as the presence of cars and pedestrians in such cases will limit the possibility to overtake.
4M. Paulsen et al. /Procedia Computer Science 00 (2019) 000–000
Table 1. Average travel time and congested time according to five different scenarios.
Scenario Avg. Travel Time Per Trip Avg. Congested Travel Time Per Trip
Unlimited Capacity (Desired Speed) 16.43 min 0.00 min
Actual Infrastructure (it. 0) 16.73 min 0.29 min
Actual Infrastructure (it. 200) 16.67 min 0.12 min
Single-Lane Infrastructure (it. 0) 18.76 min 2.33 min
Single-Lane Infrastructure (it. 200) 17.70 min 0.81 min
4. Results
The assignment model was run for the population and network described in Section 3 for a total of 200 iterations
with fixed choice sets after 160 iterations. The average computation time per iteration was 14.9 minutes on a single
node of a high performance computer with two 2.8 GHz deca-core processors with 120 GB RAM.
Table 1 shows the average travel time and congested travel time for five different scenarios. It is seen that in the
very first iteration, corresponding to no one adapting their routes according to the network performance, using the
actual infrastructure a cyclist would on an average get his/her average trip travel time prolonged by approximately 18
seconds of congested travel corresponding to just under 2 % of the total travel time. Due to the possibility to choose
alternative routes, this number is eventually reduced so that the average congested travel time is 7 seconds (roughly
1 %) while lowering the overall travel time by a similar amount.
This may be interpreted as bicycle congestion being a non-issue. On the other hand, it can also be seen as an
indication of how well the profound bicycle infrastructure found in Copenhagen is relieving congestion. This is
supported by the two final rows of Table 1, which are based on model runs where all links only have a single pseudo-
lane, thus radically reducing the possibility to overtake. With such an infrastructural setup, using distance based
shortest path for every cyclist results in a travel time that is prolonged by an average of 2.03 minutes per trip (12 %).
In such a scenario large benefits can be made by adapting the route, and eventually the excess travel time can be
reduced to 1.03 minutes (6.2 %) with 48 seconds of the travel time being congested time, more than six times as
much than with the actual infrastructure. It is worth noting that these numbers are average values across an area
larger than the entire Capital Region of Denmark which also covers a lot of smaller towns and rural areas where
bicycle congestion is practically non-existing. This means that the congested travel time of trips in the central parts of
Copenhagen will be considerably higher than the average values across the entire area presented in the table.
In order to investigate how the route choice is affected by congestion in the central part of Copenhagen, a differ-
ence map based on the morning peak hour from 7am to 8am between the actual congested network and a free-flow
assignment is presented in in Figure 1. At first it might be hard to see any clear patterns in the map, but including
some local knowledge lets the map be understood more easily. In the south-eastern part of the map it is seen that a
substantial amount of cyclists are avoiding Knippelsbro. More than 100 cyclists per direction deviate onto the Inner
Harbour bridge, and for the busy north-western direction the route across Langebro also becomes a reasonable alter-
native. This indeed seems plausible as Torvegade (the street linking to Knippelsbro from south-east) has a relatively
moderate bicycle infrastructure that only allows two efficient lanes per direction, although having one of the highest
daily bicycle counts. In the other end of the city centre, it is seen that Queen Louise’s Bridge is gaining cyclists due to
congestion from competing bridges. This seems intuitive as this bridge and Nørrebrogade (the street linking to Queen
Louise’s Bridge from north-west) forms a corridor that predominantly have three efficient lanes following a radical
renovation in 2011, and is considered as the main arterial road for bicycle traffic connecting westbound suburbs.
5. Conclusions & Future Work
In this study a traffic assignment model for bicycle traffic was proposed using a detailed, demand sensitive network
loading model15. A large-scale case study was conducted for the greater Copenhagen area with running times proving
that the presented methodology is in fact large-scale applicable.
The results have shown that based on the proposed model, congestion related route changes does seem to happen
in Copenhagen with some cyclists avoiding the most congested links in favour of links with particularly wide bicycle
Mads Paulsen et al. / Procedia Computer Science 151 (2019) 820–825 823
M. Paulsen et al. /Procedia Computer Science 00 (2019) 000–000 3
time, i.e. the difference between the actual travel time and the free flow travel time, is penalised 50 % harder than free
flow travel time, in line with previous findings for car users in the area17.
Each agent has a maximum number of five plans in his/her choice set, and after each iteration the performed plan
is assigned a score based on the obtained travel time in the previous iteration. New plans are added to the choice
set by performing a shortest path search in a network with empirical average travel times in time bins of 15 minutes
based on the previous iteration. However, during the shortest path search the link cost of every link is agent-specific
by using the maximum between the empirical (time binned) travel time and the free flow travel time based on that
agent’s desired speed.
Although the framework allows extension to include other link (to link) attributes previously seen in the literature
such as signals, slope, surface, and left/right turns, for the time being only travel time is used in order to focus on the
congestion caused by the network loading model.
The traffic assignment model has been implemented in the open-source and agent-based transport simulation soft-
ware MATSim16 by replacing parts of the default mobility simulator with the model from Paulsen et al. (2018)15,
alongside minor changes in the routing in order to account for individual desired speeds.
3. Case Study
In order to test the realism and the large-scale applicability of the model a case study for the greater Copenhagen
area is carried out. The demand is based on the Copenhagen Model for Person Activity Scheduling (COMPAS)18,
producing daily activity plans for a synthetic population of the area. Only persons with a bicycle trip during his/her
day have been used for this study, resulting in a population of 547,085 persons with daily activity plans forming a
total of 1,082,958 bicycle trips.
Based on aerial data of observed, uncongested bicycle traffic on Smallegade in Frederiksberg, every individual is
assigned a normally distributed headway distance parameter and a desired speed (v0in m/s) based on Johnson’s SU
distribution19 with the estimated parameters γ=−2.75, ξ=3.67, δ=4.07, and λ=3.49,15
v0∼δ
λ√2π1+v0−ξ
λ2e−1
2γ+δsinh−1v0−ξ
λ2
,(1)
inferring a mean speed and standard deviation of roughly 22 km/h and 4 km/h, respectively. This distribution has the
highest likelihood and lowest Kolmogorov-Snirnov statistic among 11 candidate distributions with estimated parame-
ters, and does not deviate significantly from the empirical distribution from Smallegade according to the Kolmogorov-
Smirnov Goodness-of-Fit test (p-value above 0.6).15
The network is based on OpenStreetMap (OSM)20 and is included in MATSim by altering the default OSM network
reader21. All link types that generally allow bicycle traffic have been included in the model unless the road explicitly
stated that bicycle traffic was not allowed. Furthermore, any road where bicycle traffic is explicitly mentioned as
allowed or designated has also been included in the network.
The network loading model15 only needs the lengths and widths of the bicycle infrastructure of links. Lengths
are always available and can always be extracted directly from the OSM data. Widths also play a vital part for the
methodology, but are rarely available. On the few links where widths (ωl) were available, the number of lanes were
determined based on the widths according to a Danish study22, that found the number of efficient lanes (Ψl) of bicycle
traffic to be determined by,
Ψl=1+ωl−0.4m
1.25 m.(2)
Widths were manually added to crucial arterial roads of Copenhagen – especially the cycle superhighways – before
extracting the data from OSM. This was done in order to secure these links being capable of handing large amount of
bicycle traffic. Where widths were not available, the type of bicycle infrastructure determined the number of lanes.
Explicit cycleways where motorised traffic is not allowed, as well as roads with bicycles riding in lanes or tracks were
given two pseudo-lanes, since practically all of such infrastructure in Denmark would be wide enough to form two
efficient lanes. Roads with no information about cycleways and footpaths where bicycle traffic is allowed were only
given a single lane, as the presence of cars and pedestrians in such cases will limit the possibility to overtake.
4M. Paulsen et al. /Procedia Computer Science 00 (2019) 000–000
Table 1. Average travel time and congested time according to five different scenarios.
Scenario Avg. Travel Time Per Trip Avg. Congested Travel Time Per Trip
Unlimited Capacity (Desired Speed) 16.43 min 0.00 min
Actual Infrastructure (it. 0) 16.73 min 0.29 min
Actual Infrastructure (it. 200) 16.67 min 0.12 min
Single-Lane Infrastructure (it. 0) 18.76 min 2.33 min
Single-Lane Infrastructure (it. 200) 17.70 min 0.81 min
4. Results
The assignment model was run for the population and network described in Section 3 for a total of 200 iterations
with fixed choice sets after 160 iterations. The average computation time per iteration was 14.9 minutes on a single
node of a high performance computer with two 2.8 GHz deca-core processors with 120 GB RAM.
Table 1 shows the average travel time and congested travel time for five different scenarios. It is seen that in the
very first iteration, corresponding to no one adapting their routes according to the network performance, using the
actual infrastructure a cyclist would on an average get his/her average trip travel time prolonged by approximately 18
seconds of congested travel corresponding to just under 2 % of the total travel time. Due to the possibility to choose
alternative routes, this number is eventually reduced so that the average congested travel time is 7 seconds (roughly
1 %) while lowering the overall travel time by a similar amount.
This may be interpreted as bicycle congestion being a non-issue. On the other hand, it can also be seen as an
indication of how well the profound bicycle infrastructure found in Copenhagen is relieving congestion. This is
supported by the two final rows of Table 1, which are based on model runs where all links only have a single pseudo-
lane, thus radically reducing the possibility to overtake. With such an infrastructural setup, using distance based
shortest path for every cyclist results in a travel time that is prolonged by an average of 2.03 minutes per trip (12 %).
In such a scenario large benefits can be made by adapting the route, and eventually the excess travel time can be
reduced to 1.03 minutes (6.2 %) with 48 seconds of the travel time being congested time, more than six times as
much than with the actual infrastructure. It is worth noting that these numbers are average values across an area
larger than the entire Capital Region of Denmark which also covers a lot of smaller towns and rural areas where
bicycle congestion is practically non-existing. This means that the congested travel time of trips in the central parts of
Copenhagen will be considerably higher than the average values across the entire area presented in the table.
In order to investigate how the route choice is affected by congestion in the central part of Copenhagen, a differ-
ence map based on the morning peak hour from 7am to 8am between the actual congested network and a free-flow
assignment is presented in in Figure 1. At first it might be hard to see any clear patterns in the map, but including
some local knowledge lets the map be understood more easily. In the south-eastern part of the map it is seen that a
substantial amount of cyclists are avoiding Knippelsbro. More than 100 cyclists per direction deviate onto the Inner
Harbour bridge, and for the busy north-western direction the route across Langebro also becomes a reasonable alter-
native. This indeed seems plausible as Torvegade (the street linking to Knippelsbro from south-east) has a relatively
moderate bicycle infrastructure that only allows two efficient lanes per direction, although having one of the highest
daily bicycle counts. In the other end of the city centre, it is seen that Queen Louise’s Bridge is gaining cyclists due to
congestion from competing bridges. This seems intuitive as this bridge and Nørrebrogade (the street linking to Queen
Louise’s Bridge from north-west) forms a corridor that predominantly have three efficient lanes following a radical
renovation in 2011, and is considered as the main arterial road for bicycle traffic connecting westbound suburbs.
5. Conclusions & Future Work
In this study a traffic assignment model for bicycle traffic was proposed using a detailed, demand sensitive network
loading model15. A large-scale case study was conducted for the greater Copenhagen area with running times proving
that the presented methodology is in fact large-scale applicable.
The results have shown that based on the proposed model, congestion related route changes does seem to happen
in Copenhagen with some cyclists avoiding the most congested links in favour of links with particularly wide bicycle
824 Mads Paulsen et al. / Procedia Computer Science 151 (2019) 820–825
M. Paulsen et al. /Procedia Computer Science 00 (2019) 000–000 5
Flow Difference
]-∞ ,-100]
]-100,-10 ]
]-10 , 0 [
[ 0 , 0 ]
] 0 , 10 [
[ 10 , 100[
[ 100, ∞ [
0 10.5
km
±
Queen Louise's
Bridge
Langebro
Knippelsbro
Inner Harbour
Bridge
Fig. 1. Map with selected points of interests showing the difference in flow in the morning peak hour from 7 to 8 between modelled selected routes
and routes found by using a free-flow, distance-based assignment.
tracks. The results also showed that having a comprehensive, decent bicycle infrastructure can greatly reduce the
travel time of cyclists. However, this effect is likely overestimated with the current study as the single-lane scenario
implies no opportunity to overtake altogether. In reality cyclists do tend to find a way of overtaking, why the predicted
travel times will probably be pessimistic for the fastest cyclists.
Even though the travel time gains are possibly overestimated, the study only includes travel time as a parameter. In
reality, better infrastructure is inherently attractive for cyclists1, and will also improve the perceived safety of cyclists.
Parameters that this study do not deal with explicitly, although it would be interesting to include in future research.
Furthermore, the methodology applied in this study only deals with delays on the links themselves. However, the
majority of excess travel time for cyclists is possibly induced at intersections. In the morning peak hour certain core
links in Copenhagen have bicycle queues that are often longer than what can be emptied in a single signal cycle.
Delays can also happen when cyclists are waiting for through-going traffic when making a left turn.
Such delays can be included directly in the mobility simulation by concurrently simulating motor vehicles and
modelling right of way and signals at intersections. This would also capture how cyclists delay the remainder of
traffic, meaning that it would be very suitable for project appraisal purposes. Although relevant, implementing such
methodology would be a serious extension and possibly be on the limit on what is computationally feasible.
A pragmatic alternative could possibly be to include the number of turns and signals as parameters in the route
choice model23 and/or to apply fixed travel time penalties for every signal, left turn, and right turn, respectively.
The literature on bicycle route choice modelling also contains additional link-specific parameters such as land-use14,
surface14, and slope24, which straightforwardly can and should be added to the model in future work, although slope
is not a huge issue in the greater Copenhagen area.
6M. Paulsen et al. /Procedia Computer Science 00 (2019) 000–000
References
1. Antonakos. Environmental and travel preferences of cyclists. Transportation Research Record 1994;(1438):25–33. URL: https://trid.
trb.org/view/413764.
2. Li, Wang, Liu, Ragland. Physical environments influencing bicyclists’ perception of comfort on separated and on-street bicycle facili-
ties. Transportation Research Part D: Transport and Environment 2012;17(3):256–261. URL: http://linkinghub.elsevier.com/
retrieve/pii/S1361920911001556. doi:10.1016/j.trd.2011.12.001.
3. Bai, Liu, Chan, Li. Estimating level of service of mid-block bicycle lanes considering mixed traffic flow. Transportation Research Part A:
Policy and Practice 2017;101:203–217. URL: https://linkinghub.elsevier.com/retrieve/pii/S0965856416300374. doi:10.
1016/j.tra.2017.04.031.
4. Sheffi.Urban transportation networks: Equilibrium analysis with mathematical programming methods. 1. Englewood Cliffs, NJ,
USA: Prentice-Hall; 1985. ISBN 0139397299. URL: http://web.mit.edu/sheffi/www/selectedMedia/sheffi_urban_trans_
networks.pdf.
5. Agarwal, Ziemke, Nagel. Bicycle superhighway: An environmentally sustainable policy for urban transport. VSP Working Paper 17-
16, TU Berlin, Transport Systems Planning and Transport Telematics; 2017. URL: https://www.researchgate.net/publication/
320136429_Bicycle_superhighway_An_environmentally_sustainable_policy_for_urban_transport.
6. Pritchard. Revealed Preference Methods for Studying Bicycle Route Choice—A Systematic Review. International Journal of Environmental
Research and Public Health 2018;15(3):470. URL: http://www.mdpi.com/1660-4601/15/3/470. doi:10.3390/ijerph15030470.
7. Hood, Sall, Charlton. A GPS-based bicycle route choice model for San Francisco, California. Transportation Letters 2011;3(1):63–75. URL:
http://www.tandfonline.com/doi/full/10.3328/TL.2011.03.01.63-75. doi:10.3328/TL.2011.03.01.63-75.
8. Zimmermann, Mai, Frejinger. Bike route choice modeling using GPS data without choice sets of paths. Transportation Research Part C:
Emerging Technologies 2017;75:183–196. URL: https://linkinghub.elsevier.com/retrieve/pii/S0968090X16302637. doi:10.
1016/j.trc.2016.12.009.
9. Ehrgott, Wang, Raith, van Houtte. A bi-objective cyclist route choice model. Transportation Research Part A: Policy and Practice 2012;
46(4):652–663. URL: http://linkinghub.elsevier.com/retrieve/pii/S0965856411001844. doi:10.1016/j.tra.2011.11.
015.
10. Jacyna, Wasiak, Kłodawski, Gołe¸biowski. Modelling of Bicycle Traffic in the Cities Using VISUM. Procedia Engineering 2017;187:435–
441. URL: https://www.sciencedirect.com/science/article/pii/S1877705817319276. doi:10.1016/j.proeng.2017.04.
397.
11. Ziemke, Metzler, Nagel. Modeling bicycle traffic in an agent-based transport simulation. Procedia Computer Science 2017;109(00):923–928.
URL: https://linkinghub.elsevier.com/retrieve/pii/S1877050917311043. doi:10.1016/j.procs.2017.05.424.
12. Ryu, Chen, Su, Choi. Two-Stage Bicycle Traffic Assignment Model. Journal of Transportation Engineering, Part A: Systems 2018;
144(2):04017079. URL: http://ascelibrary.org/doi/10.1061/JTEPBS.0000108. doi:10.1061/JTEPBS.0000108.
13. Ziemke, Metzler, Nagel. Bicycle traffic and its interaction with motorized traffic in an agent-based transport simulation framework. Fu-
ture Generation Computer Systems 2018;URL: https://linkinghub.elsevier.com/retrieve/pii/S0167739X17320447. doi:10.
1016/j.future.2018.11.005.
14. Prato, Halld´
orsd´
ottir, Nielsen. Evaluation of land-use and transport network effects on cyclists’ route choices in the Copenhagen Region
in value-of-distance space. International Journal of Sustainable Transportation 2018;:1–12URL: https://www.tandfonline.com/doi/
full/10.1080/15568318.2018.1437236. doi:10.1080/15568318.2018.1437236.
15. Paulsen, Rasmussen, Nielsen. Fast or Forced to Follow: A Speed-Heterogeneous Approach to Congested Multi-Lane Bicycle Traffic Simu-
lation. Manuscript submitted for publication; 2018.
16. Horni, Nagel, Axhausen, editors. The Multi-Agent Transport Simulation MATSim. London: Ubiquity Press; 2016. URL: https://www.
ubiquitypress.com/site/books/10.5334/baw/. doi:10.5334/baw.
17. Prato, Rasmussen, Nielsen. Estimating Value of Congestion and of Reliability from Observation of Route Choice Behavior of Car Drivers.
Transportation Research Record: Journal of the Transportation Research Board 2014;2412(1):20–27. URL: http://journals.sagepub.
com/doi/10.3141/2412-03. doi:10.3141/2412-03.
18. Prato, Rasmussen, Nielsen, Watling. A disaggregate pseudo-dynamic assignment for the activity-based model of the Greater Copenhagen
Area. In: Joao, editor. 13th World Conference on Transport Research (WCTR ). Rio de Janeiro, Brazil: Federal University of Rio de Janeiro;
2013, p. 1–19.
19. Johnson. Systems of Frequency Curves Generated by Methods of Translation. Biometrika 1949;36(1/2):149–176. URL: https://www.
jstor.org/stable/2332539?origin=crossref. doi:10.2307/2332539.
20. OpenStreetMap. Accessed 13 December 2018. URL: https://www.openstreetmap.org.
21. Zilske, Neumann, Nagel. OpenStreetMap For Traffic Simulation. In: Schmidt, Gartner, editors. Proceedings of the 1st European State of the
Map – OpenStreetMap conference, no. 11-10. 2011, p. 126–134.
22. Buch, Greibe. Analysis of Bicycle Traffic on One-Way Bicycle Tracks of Different Width. In: European Transport Conference 2015.
Frankfurt, Germany: Association for European Transport (AET); 2015, URL: https://aetransport.org/en-gb/past-etc-papers/
conference-papers-2015?abstractId=4418&state=b.
23. Broach, Dill, Gliebe. Where do cyclists ride? A route choice model developed with revealed preference GPS data. Transporta-
tion Research Part A: Policy and Practice 2012;46(10):1730–1740. URL: http://linkinghub.elsevier.com/retrieve/pii/
S0965856412001164. doi:10.1016/j.tra.2012.07.005.
24. Menghini, Carrasco, Sch¨
ussler, Axhausen. Route choice of cyclists in Zurich. Transportation Research Part A: Policy and Practice
2010;44(9):754–765. URL: http://linkinghub.elsevier.com/retrieve/pii/S0965856410001187. doi:10.1016/j.tra.2010.
07.008.
Mads Paulsen et al. / Procedia Computer Science 151 (2019) 820–825 825
M. Paulsen et al. /Procedia Computer Science 00 (2019) 000–000 5
Flow Difference
]-∞ ,-100]
]-100,-10 ]
]-10 , 0 [
[ 0 , 0 ]
] 0 , 10 [
[ 10 , 100[
[ 100, ∞ [
0 10.5
km
±
Queen Louise's
Bridge
Langebro
Knippelsbro
Inner Harbour
Bridge
Fig. 1. Map with selected points of interests showing the difference in flow in the morning peak hour from 7 to 8 between modelled selected routes
and routes found by using a free-flow, distance-based assignment.
tracks. The results also showed that having a comprehensive, decent bicycle infrastructure can greatly reduce the
travel time of cyclists. However, this effect is likely overestimated with the current study as the single-lane scenario
implies no opportunity to overtake altogether. In reality cyclists do tend to find a way of overtaking, why the predicted
travel times will probably be pessimistic for the fastest cyclists.
Even though the travel time gains are possibly overestimated, the study only includes travel time as a parameter. In
reality, better infrastructure is inherently attractive for cyclists1, and will also improve the perceived safety of cyclists.
Parameters that this study do not deal with explicitly, although it would be interesting to include in future research.
Furthermore, the methodology applied in this study only deals with delays on the links themselves. However, the
majority of excess travel time for cyclists is possibly induced at intersections. In the morning peak hour certain core
links in Copenhagen have bicycle queues that are often longer than what can be emptied in a single signal cycle.
Delays can also happen when cyclists are waiting for through-going traffic when making a left turn.
Such delays can be included directly in the mobility simulation by concurrently simulating motor vehicles and
modelling right of way and signals at intersections. This would also capture how cyclists delay the remainder of
traffic, meaning that it would be very suitable for project appraisal purposes. Although relevant, implementing such
methodology would be a serious extension and possibly be on the limit on what is computationally feasible.
A pragmatic alternative could possibly be to include the number of turns and signals as parameters in the route
choice model23 and/or to apply fixed travel time penalties for every signal, left turn, and right turn, respectively.
The literature on bicycle route choice modelling also contains additional link-specific parameters such as land-use14,
surface14, and slope24, which straightforwardly can and should be added to the model in future work, although slope
is not a huge issue in the greater Copenhagen area.
6M. Paulsen et al. /Procedia Computer Science 00 (2019) 000–000
References
1. Antonakos. Environmental and travel preferences of cyclists. Transportation Research Record 1994;(1438):25–33. URL: https://trid.
trb.org/view/413764.
2. Li, Wang, Liu, Ragland. Physical environments influencing bicyclists’ perception of comfort on separated and on-street bicycle facili-
ties. Transportation Research Part D: Transport and Environment 2012;17(3):256–261. URL: http://linkinghub.elsevier.com/
retrieve/pii/S1361920911001556. doi:10.1016/j.trd.2011.12.001.
3. Bai, Liu, Chan, Li. Estimating level of service of mid-block bicycle lanes considering mixed traffic flow. Transportation Research Part A:
Policy and Practice 2017;101:203–217. URL: https://linkinghub.elsevier.com/retrieve/pii/S0965856416300374. doi:10.
1016/j.tra.2017.04.031.
4. Sheffi.Urban transportation networks: Equilibrium analysis with mathematical programming methods. 1. Englewood Cliffs, NJ,
USA: Prentice-Hall; 1985. ISBN 0139397299. URL: http://web.mit.edu/sheffi/www/selectedMedia/sheffi_urban_trans_
networks.pdf.
5. Agarwal, Ziemke, Nagel. Bicycle superhighway: An environmentally sustainable policy for urban transport. VSP Working Paper 17-
16, TU Berlin, Transport Systems Planning and Transport Telematics; 2017. URL: https://www.researchgate.net/publication/
320136429_Bicycle_superhighway_An_environmentally_sustainable_policy_for_urban_transport.
6. Pritchard. Revealed Preference Methods for Studying Bicycle Route Choice—A Systematic Review. International Journal of Environmental
Research and Public Health 2018;15(3):470. URL: http://www.mdpi.com/1660-4601/15/3/470. doi:10.3390/ijerph15030470.
7. Hood, Sall, Charlton. A GPS-based bicycle route choice model for San Francisco, California. Transportation Letters 2011;3(1):63–75. URL:
http://www.tandfonline.com/doi/full/10.3328/TL.2011.03.01.63-75. doi:10.3328/TL.2011.03.01.63-75.
8. Zimmermann, Mai, Frejinger. Bike route choice modeling using GPS data without choice sets of paths. Transportation Research Part C:
Emerging Technologies 2017;75:183–196. URL: https://linkinghub.elsevier.com/retrieve/pii/S0968090X16302637. doi:10.
1016/j.trc.2016.12.009.
9. Ehrgott, Wang, Raith, van Houtte. A bi-objective cyclist route choice model. Transportation Research Part A: Policy and Practice 2012;
46(4):652–663. URL: http://linkinghub.elsevier.com/retrieve/pii/S0965856411001844. doi:10.1016/j.tra.2011.11.
015.
10. Jacyna, Wasiak, Kłodawski, Gołe¸biowski. Modelling of Bicycle Traffic in the Cities Using VISUM. Procedia Engineering 2017;187:435–
441. URL: https://www.sciencedirect.com/science/article/pii/S1877705817319276. doi:10.1016/j.proeng.2017.04.
397.
11. Ziemke, Metzler, Nagel. Modeling bicycle traffic in an agent-based transport simulation. Procedia Computer Science 2017;109(00):923–928.
URL: https://linkinghub.elsevier.com/retrieve/pii/S1877050917311043. doi:10.1016/j.procs.2017.05.424.
12. Ryu, Chen, Su, Choi. Two-Stage Bicycle Traffic Assignment Model. Journal of Transportation Engineering, Part A: Systems 2018;
144(2):04017079. URL: http://ascelibrary.org/doi/10.1061/JTEPBS.0000108. doi:10.1061/JTEPBS.0000108.
13. Ziemke, Metzler, Nagel. Bicycle traffic and its interaction with motorized traffic in an agent-based transport simulation framework. Fu-
ture Generation Computer Systems 2018;URL: https://linkinghub.elsevier.com/retrieve/pii/S0167739X17320447. doi:10.
1016/j.future.2018.11.005.
14. Prato, Halld´
orsd´
ottir, Nielsen. Evaluation of land-use and transport network effects on cyclists’ route choices in the Copenhagen Region
in value-of-distance space. International Journal of Sustainable Transportation 2018;:1–12URL: https://www.tandfonline.com/doi/
full/10.1080/15568318.2018.1437236. doi:10.1080/15568318.2018.1437236.
15. Paulsen, Rasmussen, Nielsen. Fast or Forced to Follow: A Speed-Heterogeneous Approach to Congested Multi-Lane Bicycle Traffic Simu-
lation. Manuscript submitted for publication; 2018.
16. Horni, Nagel, Axhausen, editors. The Multi-Agent Transport Simulation MATSim. London: Ubiquity Press; 2016. URL: https://www.
ubiquitypress.com/site/books/10.5334/baw/. doi:10.5334/baw.
17. Prato, Rasmussen, Nielsen. Estimating Value of Congestion and of Reliability from Observation of Route Choice Behavior of Car Drivers.
Transportation Research Record: Journal of the Transportation Research Board 2014;2412(1):20–27. URL: http://journals.sagepub.
com/doi/10.3141/2412-03. doi:10.3141/2412-03.
18. Prato, Rasmussen, Nielsen, Watling. A disaggregate pseudo-dynamic assignment for the activity-based model of the Greater Copenhagen
Area. In: Joao, editor. 13th World Conference on Transport Research (WCTR ). Rio de Janeiro, Brazil: Federal University of Rio de Janeiro;
2013, p. 1–19.
19. Johnson. Systems of Frequency Curves Generated by Methods of Translation. Biometrika 1949;36(1/2):149–176. URL: https://www.
jstor.org/stable/2332539?origin=crossref. doi:10.2307/2332539.
20. OpenStreetMap. Accessed 13 December 2018. URL: https://www.openstreetmap.org.
21. Zilske, Neumann, Nagel. OpenStreetMap For Traffic Simulation. In: Schmidt, Gartner, editors. Proceedings of the 1st European State of the
Map – OpenStreetMap conference, no. 11-10. 2011, p. 126–134.
22. Buch, Greibe. Analysis of Bicycle Traffic on One-Way Bicycle Tracks of Different Width. In: European Transport Conference 2015.
Frankfurt, Germany: Association for European Transport (AET); 2015, URL: https://aetransport.org/en-gb/past-etc-papers/
conference-papers-2015?abstractId=4418&state=b.
23. Broach, Dill, Gliebe. Where do cyclists ride? A route choice model developed with revealed preference GPS data. Transporta-
tion Research Part A: Policy and Practice 2012;46(10):1730–1740. URL: http://linkinghub.elsevier.com/retrieve/pii/
S0965856412001164. doi:10.1016/j.tra.2012.07.005.
24. Menghini, Carrasco, Sch¨
ussler, Axhausen. Route choice of cyclists in Zurich. Transportation Research Part A: Policy and Practice
2010;44(9):754–765. URL: http://linkinghub.elsevier.com/retrieve/pii/S0965856410001187. doi:10.1016/j.tra.2010.
07.008.
