A door-to-door route optimizer that includes sustainability criteria

FINAL DEGREE PROJECT TITLE: A door-to-door route optimizer that includes sustainability criteria DEGREE: Bachelor’s Degree in Aerospace Systems Engineering (major in Air Navigation) AUTHOR: Carlos Jaime Hermoso DIRECTOR: Adeline de Villardi de Montlaur DATE: February 8th 2021 Resumen Título: Un optimizador de rutas puerta a puerta que incluye criterios de sostenibilidad Autor: Carlos Jaime Hermoso Director: Adeline de Villardi de Montlaur Fecha: 8 de febrero del 2021 En este proyecto se expone la metodología utilizada para desarrollar un software capaz de predecir el medio de transporte óptimo para una ruta puerta a puerta. Los criterios utilizados para comparar los diferentes medios de transporte son: el precio del viaje, el tiempo total de la ruta, la sensación de confort que transmite cada sistema de transporte y la huella ecológica que provoca realizar el viaje. A la hora de escoger la mejor alternativa, el software tiene en cuenta los pesos que se le otorgan a cada criterio, y a su vez, estos pesos se pueden repartir en función de las preferencias del usuario. De esta forma la solución óptima cambia según la importancia que pueda tener cada criterio para un tipo de usuario u otro. Una vez operativo, se utiliza el software para averiguar cuales son los mejores medios de transporte para ciertas rutas específicas. También se emplea para entender qué efecto tiene en las soluciones un escenario donde solo importa el coste y el precio del viaje, respecto a un escenario en el que se introducen criterios medioambientales. Finalmente también se aprovecha el funcionamiento del software para descubrir cómo varían las mejores soluciones de medios de transporte según usuarios con distintos perfiles. Overview Title: A door-to-door route optimizer that includes sustainability criteria Author: Carlos Jaime Hermoso Director: Adeline de Villardi de Montlaur Date: February 8th 2021 In this project, the methodology used to develop a software capable of predicting the optimal means of transport for a door-to-door route is exposed. The criteria used to compare the different means of transport are: the price of the trip, the total time of the route, the feeling of comfort that each transport system transmits and the ecological footprint that the trip causes. When choosing the best alternative, the software takes into account the weights that are assigned to each criterion, and in turn, these weights can be distributed according to the user's preferences. In this way, the optimal solution changes according to the importance that each criterion may have for one type of user or another. Once operational, the software is used to find out the best means of transportation for specific routes. It is also employed to understand what effect a scenario where only the cost and price of the trip has on the solutions, compared to a scenario in which environmental criteria are introduced. Finally, the operation of the software is also exploited to discover how the best transportation solutions vary according to users with different profiles. CONTENTS LIST OF FIGURES I ............................................................................................... LIST OF TABLES II ................................................................................................ LIST OF ABBREVIATIONS III ............................................................................... INTRODUCTION IV ................................................................................................ I. Motivation IV ................................................................................................................ II. Goals IV ........................................................................................................................ III. Outline IV ..................................................................................................................... IV. Acknowledgments V ................................................................................................... 1. TRANSPORT DATA 1 ...................................................................................... 1.1. Car 1 ................................................................................................................................ 1.1.1. Distance 1 ................................................................................................... 1.1.2. Time 2 .......................................................................................................... 1.1.3. Price 2 ......................................................................................................... 1.1.4. Footprint 5 .................................................................................................. 1.2. Airplane 6 ........................................................................................................................ 1.2.1. Distance 6 ................................................................................................... 1.2.2. Time 7 .......................................................................................................... 1.2.3. Price 10 ....................................................................................................... 1.2.4. Footprint 10 ................................................................................................ 1.3. Train 11 ............................................................................................................................ 1.3.1. Distance 11 ................................................................................................. 1.3.2. Time 11 ........................................................................................................ 1.3.3. Price 12 ....................................................................................................... 1.3.4. Footprint 12 ................................................................................................ 1.4. Energy resources 13 ................................................................................................... 1.5. Comfort 14 ................................................................................................................... 2. MULTI-CRITERIA DECISION MAKING 17 ..................................................... 2.1. VIKOR method 18 ............................................................................................................... 2.2. Weights 21 .......................................................................................................................... 3. SOFTWARE 26 ................................................................................................ 3.1. Inputs 26 ............................................................................................................................. 3.2. Searches 27 ........................................................................................................................ 3.3. Calculations 27 ................................................................................................................... 3.4. Matrix 27 ............................................................................................................................. 3.5. VIKOR 28 ............................................................................................................................. 3.6. Outputs 28 .......................................................................................................................... 3.7. UML Diagram 29 ................................................................................................................. 3.8. Classes 31 .......................................................................................................................... 4. RESULTS 35 .................................................................................................... 4.1. Cost vs time scenario 38 ................................................................................................... 4.2. Cost vs time vs emission balanced scenario 40 ............................................................. 4.3. Multi criteria user based scenario 42 ............................................................................... 4.4. Comparison of the results of the scenarios 47 ............................................................... 5. CONCLUSIONS 48 .......................................................................................... 5.1. Future work 48 ................................................................................................................... REFERENCES 50 .................................................................................................. ANNEX A. DATA MODELING OF THE DIFFERENT CRITERIA USED TO COMPARE MEANS OF TRANSPORT 54 ............................................................. Annex A.1 Model for calculating travel time by car in relation to distance 54 ........................ Annex A.2 Extra cost vs Saved time, compared to cheaper but slower routes 54 ................. Annex A.3 Price of fuel and tolls against distance according to national routes. 55 ............. Annex A.4 Model for calculating footprint by car in relation to distance 55 ........................... Annex A.5 Amount of CO2 proportional to a passenger per km (gCO2 / RPK) 56 ................. Annex A.6 Behavior of the train ticket price according to the search dates 57 ...................... Annex A.7 Results of the comfort survey 58 .............................................................................. Annex A.8 Comfort trends based on survey responses to four cities with different distances 62 ................................................................................................................................... ANNEX B. EXAMPLE OF THE RESPONSE IN JSON FORMAT OF A REQUEST TO THE SKYSCANNER API 64 .......................................................... ANNEX C. RESULTS TABLES IN THE THIRD EXPERIMENT SCENARIO 65.... I LIST OF FIGURES •Fig. 1.1 Price of fuel (blue) and tolls (green) against distance according to searches from Castelldefels-Barcelona in [Via Michelin] for a “Citroën C4 1.6L 110 CV”. Gasoil price: 1.16 €. International routes •Fig. 1.2 Trend line for fuel price vs distance according to searches from Barcelona in [Via Michelin] for a Citroën C4 1.6L 110 CV. Gasoil price: 1.16 € •Fig. 1.3 Trend line for fuel price plus toll price vs distance according to searches from Barcelona in [Via Michelin] for a Citroën C4 1.6L 110CV. Gasoil price: 1.16€ •Fig. 1.4 Departure Delay Causes in 2020 and 2019 according to Eurocontrol [Eurocontrol, 2020] •Fig. 1.5 Flight time versus flight distance according to searches on FlightRadar24 [FlightRadar24] •Fig. 1.6 Energy resources needed against distance for car, airplane and train •Fig. 1.7 Answers to the question: "What means of transport do you consider most comfortable to go to the following destinations?” •Fig. 2.1 Weights of different groups of people •Fig. 3.1 Steps of optimization software •Fig. 3.2 UML Diagram of the software •Fig. 4.1 values in the six routes from Castelldefels for a “Two-Criteria” scenario •Fig. 4.2 values in the six routes from Castelldefels for a “Balanced” scenario •Fig. 4.3 Number of times that each means of transport has been the best solution among the 6 groups •Fig. 4.4 Number of times that each means of transport has been the worst solution among the 6 groups •Fig. 4.5 Average value for the six groups in the six routes from Castelldefels •Fig. 4.6 Average value for the six groups Q Q Q Q 3 research of this project, the free resources offered by ViaMichelin have been used to try to model the cost of car trips. Via Michelin's official page allows you to use route search resources for free. The procedure carried out to estimate price has been the same as for estimating time, through a linear regression (Section 1.1.2 and Annex A.1). That is, various routes have been searched, and the distance, fuel price and toll price data have been recorded. In order for the data of the study in section 4 to be as accurate as possible, it has been decided that the routes to be analyzed will all be located in Castelldefels (Barcelona) because this will be the starting location for the experiments. In Fig 1.1 it can be seen for some of the selected routes (routes with international destinations), highlighted in blue the cost of fuel and in green the tolls cost. The vehicle used for these routes is a Citroen C4 1.6 L 110 CV and the Diesel Price at the time of the search was € 1.16. Fig. 1.1 Price of fuel (blue) and tolls (green) against distance according to searches from Castelldefels-Barcelona in [Via Michelin] for a “Citroën C4 1.6L 110 CV”. Gasoil price: 1.16 €. International routes Price (€) 0 87,5 175 262,5 350 Distance (Km) 667 1004 1196 1369 1743 Fuel Cost (€) Toll Cost (€) A door-to-door route optimizer that includes sustainability criteria 4 Looking at the blue bars of Fig 1.1, it can be seen how the price of fuel increases with the distance of the trip, making it possible to approximate it with a linear regression, since the consumption of the car is directly proportional to the distance (Fig 1.2). On the other hand, the cost of tolls (green bars in Fig 1.1) has a greater difficulty to predict, since its dependence in this case is not with the distance but with the destination of the route and the countries it crosses. Fig. 1.2 Trend line for fuel price vs distance according to searches from Barcelona in [Via Michelin] for a Citroën C4 1.6L 110 CV. Gasoil price: 1.16 € Via Michelin offers its users the possibility of knowing the cheapest route and the fastest route. In some routes the result of the two of them was the same or it changed just a little bit to select one road over another. But in other cases, there was a difference between picking the route with the least tolls or the fastest, regardless of the price. These cases have been taken into account in the investigation. In Annex A.2 it can be seen that the relationship between the time gained for choosing the fastest route and the price paid in tolls has no relationship, this being a very different value according to each case. Another of the patterns found in the research is, that within the searches for national routes, it is easier to find routes without tolls (regardless of the extra time). However, in most international routes they require mandatory tolls (for example, the “vignette” in Switzerland of approximately 40 € per year, compulsory if you enter Swiss territory) that increase the cost of journey making it more difficult to find a slower but cheaper route. In Annex A.3 it can be seen the price of fuel and tolls against distance according to national routes. Fuel Cost (€) 0 70 140 210 280 350 Distance (Km) 0 250 500 750 1000 1250 1500 1750 2000 y = 0,109x + 3,0994 R² = 0,9949 5 Fig. 1.3 Trend line for fuel price plus toll price vs distance according to searches from Barcelona in [Via Michelin] for a Citroën C4 1.6L 110CV. Gasoil price: 1.16€ In Fig 1.3, the total cost (fuel + toll) of traveling by car is modeled with a regression line for all the cases used in the research. As it has been explained, there are several alternatives that can make the trip more expensive or cheaper on a voluntary basis (choosing one route or another) or involuntarily (when crossing through certain countries). The trend line of all cases would have a coefficient of determination of 0.81, caused almost entirely by the large variation in the price of the toll in each case. For this reason, for the experiments in section 4, we will use routes from Castelldefels with real data on toll and fuel costs, such as those found in this section, in order to improve the precision of our experiments. For the rest of the searches carried out with the software, the Total cost vs Distance model in Fig 1.3 will be used. 1.1.4. Footprint To calculate the footprint of the car trip, the value of carbon dioxide emissions emitted by the vehicle during the trip will be used. As with the travel time, the consumption of a vehicle on the highway is usually constant, as it is proportional to the speed of the vehicle. As most of the trip takes place at a constant speed, it is effective to estimate CO2 emissions based on the distance calculated in section 1.1.1. To make this linear model that relates emissionsdistance, two external resources have been used. First of all, the services of Via Michelin are used again. All the searches used to find out the price in section 1.1.3 also include information on the carbon dioxide emissions emitted. On the other side, the EcoPassengers emissions calculator has also been used [Ifeu, 2010]. This service is better defined in section 1.3.4 where the model used for Total Cost vs Distance Total Cost (€) 0 70 140 210 280 350 Distance (Km) 0 250 500 750 1000 1250 1500 1750 2000 y = 0,1756x + 5,2576 R² = 0,8182 A door-to-door route optimizer that includes sustainability criteria 6 train emissions is explained as the EcoPassengers emissions calculator is developed in cooperation with International Union of Railways (UIC) [UIC]. In any case, it also has a solid scientific methodology to determine data from other means of transport than the train. Thanks to the information from these two sources, it has been possible to collect enough data to create our own model. After putting together several results in a scatter plot, see Annex A.4, a trend line is obtained with a coefficient of determination of 0.9975. The resulting equation is equation (1.1) (1.1) Where is footprint calculated in Kg of CO2 and is distance in Km. 1.2. Airplane One of the main goals of this project is to be able to compare the movement from point A to point B through different kinds of transport systems in the most precise way possible. This precision, as will be discussed throughout the project, may have limitations due to different factors such as lack of resources or difficulties generating a standard model to obtain data through another variable. In order to calculate the data through air transport as accurately as possible, a series of steps has been carried out that allow the user to locate the closest airport to the two search points. When making the air route calculations in the software, the time, price and footprint of the plane trip are taken into account. But also, the movement by vehicle from the Initial location chosen by the user to the departure airport, and from the arrival airport to the destination location are added to the final data. This section only explains the procedure to obtain the air travel data, since the resources of section 1.1 are used for the calculations of routes from the main locations to the respective airports. 1.2.1. Distance The distance between airports in a straight line will be necessary to model the travel time and the emitted footprint. To calculate the distance in a straight line, the Haversine formula (1.2) has been used (1.2) Where is the equivolume radius of the Earth (6371 Km) and: fcar = 0.2161 dcar + 5.3715 fcar dcar d=R c R 7 (1.3) Where is: (1.4) Where: (1.5) (1.6) The inputs , , , correspond to the latitudes and longitudes of the origin airport and destination airport respectively. These can be obtained from the response JSON of the Google Directions API, therefore, when calculating the routes by car to the respective airports, the coordinates of both will be used to calculate through this method the distance in a straight line that separates it. 1.2.2. Time It is difficult to predict the exact time of a plane trip since there are several facts that can modify the departure time and the pre-established arrival time of a flight. Looking at the Eurocontrol records [Eurocontrol, 2020] we can see that in November the average departure delay was 5.8 minutes per flight and the arrival delay was 4.8 minutes per flight. These figures are lower than the ones on 2019: 8.4 minutes per flight and 7.3 minutes per flight respectively due to the effect of the COVID-19 global pandemic, which has caused the air traffic network to decline by 61.6% compared to November of the previous year. Following the data from All-Causes Delay to Air Transport in Europe November 2019 and taking this 2019 data as a reference (to avoid working with the exceptional data of 2020 due the global pandemic), the average delay in departure for each flight is divided into the different causes that appear in Fig 1.4 c= 2 atan2( a, (1 −a)) a a= sin2(Δlat/2) + cos(lat1) cos(lat2) sin2(Δlong/2) Δlat =lat2−lat1 Δlong =long2−long1 lat1 lat2 long1 long2 A door-to-door route optimizer that includes sustainability criteria 8 Fig. 1.4 Departure Delay Causes in 2020 and 2019 according to Eurocontrol [Eurocontrol, 2020] Saving the reasons mentioned above that can cause a delay in the take-off or landing of a flight, modifying your Estimated time of departure (ETD) and Estimated time of Arrival (ETA), the exact travel time between one city and another is usually quite a lot similar. The exact duration of a flight cannot be predicted due to Air Traffic Flow Management (ATFM) or weather reasons for example, but for this work it has been considered enough to approximate the distance-time relationship by means of a linear approximation. In order to make this possible, historical data of several flights have been collected from the official FlightRadar24 page [FlightRadar24]. For each flight, the duration of the last 7 times that flight has traveled the same route has been averaged. Considering flights of very short distances to long distances, the result can be seen on Fig 1.5 Average Delay / Flights (mins) 0 1,5 3 4,5 6 7,5 9 Departure Delay Causes Reactionary Airline ATFM En Route ATFM Airport Other Weather Government Miscellaneous ATFM Weather Other Airport TOTAL 2020 2019 9 Fig. 1.5 Flight time versus flight distance according to searches on FlightRadar24 [FlightRadar24] In the graph of Fig 1.5 it is shown how as soon the distance increases, the deviations from the average line are greater. This variation depends mainly on the orientation of the route. Due to the Polar Jet Stream in the northern hemisphere [NASA, 2011] accelerated air currents that follow the rotation, the London - New York flight can last an hour longer from the earth and that are caused by the convergence of cold air descending from the Arctic together with masses of warm air from the tropics. For this reason, to approximate the travel time following a linear function, it has been considered appropriate to divide it into three different sections. A single approach for flights less than 4000 km and two different approaches for flights over 4000 km with a heading, following the rotation of the earth and with an adverse heading. However, for the experiments of this project, the distances of the routes are all less than 2000 km since the objective is to compare the alternatives of airplane with those of train and car in situations in which they can compete. For distances much greater than 2000 km it is quite difficult to imagine a scenario in which the airplane is not the best option for the trip. For this reason, the equation used for the experiments of this project is equation (1.7) ; (1.7) tair = 0.0675 dair + 22.369 {100 < dair < 2000} Flight time vs Distance Flight time (min) 0 125 250 375 500 625 750 875 1000 Distance (km) 0 1750 3500 5250 7000 8750 10500 12250 14000 y = 0,0675x + 22,369 R² = 0,9903 A door-to-door route optimizer that includes sustainability criteria 10 Where is the flight time in minutes and is the distance between airports in km. The formula, apart from being limited to greater distances, is also limited to distances less than 100 km to avoid impossible scenarios 1.2.3. Price The flight market is somewhat fluctuating and adapts to the laws of supply and demand [El Mundo, 2013]. Factors such as aircraft capacity, historical demand, aircraft model, frequencies for that route, competition or anticipation of purchase, among others, are studied daily by the 'Revenue management' departments. Specific software for optimization to decide what price can offer for the request that an end customer or travel agency is making at that time [El Mundo, 2013]. For this reason, approximating the price of the flight in a static way, such as a linear approximation, has been done to estimate the time regarding the distance and the heading of the flight in section 1.2.1 it is more complicated. Luckily thanks to RapidAPI it has been possible to have free access to the SkyScanner API [SkyScanner API] which allows you to know the price of a flight for a specific day at the time the search is being carried out. This is something very interesting since it turns the project software into a non-static program, so that the optimal route results can vary depending on the day the experiments are carried out or the day the trip is to be made. The SkyScanner API allows requests to be made automatically with the software and the API returns a JSON from which the price information between the two airports can be extracted. 1.2.4. Footprint The global aviation industry produces some more than 2% (781 million tonnes) of all human CO2 emissions (36 trillion tonnes in 2015) also Aviation is responsible for 12% of CO2 emissions from all transport sources, compared to 74% from road transport [Europarl, 2019]. In section 1.1.4, it was explained how carbon dioxide emissions for a car trip can be modeled through a linear expression because consumption for most of the trip is usually constant. This also happens for the plane, but only for long distances. When it comes to short distances, CO2 emissions are much higher than for long distances and it cannot be modeled with the same equation. This value is not constant because the highest intensity of carbon emitted by an aircraft occurs in the takeoff and landing phases. This reason causes that, for short distances, the kg of CO2 emitted per km are higher because although the km of travel are not high, the take-off and landing emissions continue to have a great weight. Once the approximately 2,500 km of travel has been exceeded, the emissions from cruise speed, takeoff and landing stabilize and the relationship between distance and footprint begins to be linear as the distance value increases [ICCT, 2019]. In Annex A.5 you can see the graph of [ICCT, 2019] that determines the amount of CO2 proportional to a passenger per km (gCO2 / Revenue Passenger Kilometers (RPK)) according to the distance of the trip. Using the information in tair dair 11 this graph, a model has been developed capable of calculating the total emissions of a passenger according to the distance of the trip (kgCO2 / km). The equations resulting from this model are the following: ; (1.8) ; (1.9) Where is footprint calculated in Kg of CO2 and is distance in Km. 1.3. Train The third and last means of transport used for this project is the train. The steps used to obtain the cost, time and footprint data of rail travel are explained below. 1.3.1. Distance As seen in section 1.1.1, calculating the distance by train will be done using the Google Directions API. The API allows searching for the route for rail transport including the number of transports systems necessary to get from the origin to the final destination. Therefore, the train distance that will be used in the software in this case will be the sum of the distances of all the trains necessary for this route. In this case the distance will be necessary to approximate the price and the footprint. 1.3.2. Time From the same Google Directions API’s request in section 1.3.1, the travel time is obtained for each of the journeys made on the route. Carrying out different searches, it is appreciated that the train journey between two points usually requires a greater number of stopovers if the journey is not between two large cities or, in the case of Spain, if the Alta Velocidad Española (AVE) high-speed network does not arrive. This reason makes the dependence of time on distance more difficult to achieve through a linear regression since the origin and destination locations play an important role. For this reason, for the software in section 3, the relative time values of each of the trains on the journey, the number of stopovers and the total travel time are obtained directly from the API. fair = 99.079 exp(0.1123 dair /1000) {100 < dair < 1500} fair = 42.5 dair /1000 + 127.5 {dair > 1500} fair dair A door-to-door route optimizer that includes sustainability criteria 12 1.3.3. Price Getting the price of train travel has a certain degree of complexity. As in section 1.2.3, train prices are adapted to the market and vary depending on the software and the companies experts in charge of adjusting the price. In order to estimate the value of prices trying to follow a pattern, a study has been carried out along the same routes as in section 1.1.3. Through the official website of National Network of Spanish Railways or Red Nacional de Ferrocarriles Españoles (RENFE) the price for different routes has been analyzed by comparing the price of the train ticket every day between one month and two months in advance [RENFE]. The station of origin of this study has always been Barcelona-Sants, the closest station to Castelldefels from which the vast majority of medium and long-distance trains depart. As already happened in section 1.1.3 future work wants to improve the accuracy of the software in price searches for all types of stations, for example through a payment API that works the same as the SkyScanner API, that allows us to know tickets price for free at the time of the search. But for the experiments of this project it has been deepened in knowing well the prices according to the day and the destination from Barcelona. After this procedure, the price has been determined by a linear regression, determined by the distance with a coefficient of determination of 0.94, see Equation (1.10). This study also consists of interesting results such as the behavior of the price with respect to the weeks in advance with which the search is carried out or the variation that exists between one day of the week and another. All the results of this study are attached in Annex A.6. (1.10) Where is cost calculated in € and is distance in Km. 1.3.4. Footprint The footprint calculation has been produced using the resources of EcoPassenger, a web tool that behaves like a calculator to compare energy consumption and CO2 emissions from different ways of transport. This software is developed between the UIC, the Foundation for Sustainable Development, German institute for energy and environmental research (IFEU) and HaCon (software). These calculations are fed by the UIC Environment Strategy Reporting System (ESRS) methodology [ESRS, 2016]. This calculator allows to know, among other data, the carbon dioxide emissions for a rail route between two stations. This resource has been used to take various samples and note the relationship between distance and footprint. Doing the linear regression between all the samples, equation (1.11) is obtained to approximate the footprint with the distance of the trip. ctrain = 0.0623 dtrain + 7.1893 ctrain dtrain 19 Table 2.2 Step 1: Best and worst values and The second step is to calculate the and values. These values are unique for each of the alternatives and correspond to the mathematical formulas (2.1) and (2.2) respectively. In order to calculate these values, the fixed values of Table 2.2 and the weight that each alternative receives are necessary. The weight corresponds to the percentage of relative importance that each of the criteria receives. The objective is to capture the preferences of passengers on the different criteria and convert this subjective information into percentage weights. The methodology for assigning weights used to carry out the experiments in section 4 will be detailed in section 2.2 For the explanation example of this section, the weights ( ) used are also random and correspond to the following: 30% cost, 35% time , 20% footprint, and 15% comfort. This is more important for time and cost values and less important for footprint and comfort. (2.1) (2.2) Using formulas (2.1) and (2.2) for the example, the and values of Table 2.3 corresponding to each of the alternatives are obtained. Table 2.3 Step 2: and for each alternative f* i f− i Cost (€) Time (min) Footprint (Kg CO2) Comfort (%) 50 90 20 40 90 400 150 30 f− i f* i Sj Rj wi Sj= n ∑ i=1 wi(f* i−fij)/( f* i−f− j) Rj=ma xi[wi(f* i−fij)/( fi*−fi)] Sj Rj Sj Rj Sj Rj Car 0,592 0,35 Airplane 0,65 0,3 Train 0,274 0,15 A door-to-door route optimizer that includes sustainability criteria 20 The third step will be to determine the value. The value is what allows us to rank the “closest to ideal” solutions [Pohekar and Ramachandran, 2004]. It will be done from equation (2.3) and it will be necessary to use the values of , and , that represent the maximum and minimum value of and as it happened in the first step. is used as a group maximum utility weight [1]. It takes values between 0 and 1 but normally 0.5 is used. In this example we will use 0.5 (2.3) The goal is for the results of to be 0 or as close to 0. This will determine the best solution. The ranking from best to worst alternative will be determined by values from lowest to highest. Table 2.4 shows the results of through formula (2.3) for this example of the VIKOR method. Table 2.4 Step 3: for each alternative In the case of this example, the VIKOR method has determined that for the values in Table 2.1 and with weights of: 30% cost, 35% time, 20% footprint and 15% comfort, the best solution is the train, the second best option would be the airplane and finally the car. The last step of the VIKOR method involves validation of the compromise solution through two criteria: • "Acceptable advantage": The condition (A(2) - (A(1))> = DQ is met; where A(1) is the alternative with the lowest value of , A(2) the second best alternative and DQ = 1 / (J-1). • "Acceptable Stability in decision making": Alternative A(1) apart from having the best value of must also have the minimum value of and / or . Qj Qj S* S− R* R− Sj Rj ν Qj=ν(Sj−S*)/(S−−S*) + (1 −ν)(Rj−R*)/(R−−R*) Qj Qj Qj Qj Q Car 0,923 Airplane 0,875 Train 0,0 Q Q Q Q S R 21 In the case of this example A(1) = 0.0; A(2) = 0.875 and DQ = 0.5. Therefore the first criterion is met. The values of and are 0.274 and 0.15 respectively, being also the minimum among the three alternatives. Therefore in this example the train would be an acceptable and stable optimal solution. In the applications of the VIKOR method for this project, special attention will be paid to the order of the solutions and the values of that determine the advantage that separates one solution from another. In some projects in which the problem to be solved could be similar to that of this project in which the objective is to find an optimal solution among all the alternatives according to several criteria, it has been decided to combine the VIKOR method with AHP for assigning weights [San Cristobal, 2011]. However, for the present problem in which the attributes of the alternatives are factors such as cost, time, environmental factors or comfort criteria, it has been chosen to design a specific method for this problem. The following section explains how this assignment of weights has been carried out based on the preferences of certain users. 2.2. Weights This section explains the preparation of weights for the VIKOR method. As already seen in section 1, the inputs corresponding to the three types of transportation system: car, plane and train, are the price of the trip, the total time of the route, the footprint that is issued throughout the trip and the specific comfort of each of the different means of transport. It is the moment to interpret the preferences of the users to transform them into weights for the problem. The weight for each of the inputs will be the percentage of importance, the sum of the 4 weights being 100%. It can be difficult to distribute this 100% directly among the 4 sections, so the procedure that has been carried out is to define different types of person and set specific weights for these groups. In this way, the user will not have to decide numerically the value of the weights, but must identify himself with the group of travelers that will be explained below and the software itself assigns the corresponding weights. A study by [Braintrust, 2019] tries to explain the behavior of young Spanish people when traveling. This study sectors travelers into 6 groups of people based on factors such as the average budget that the interviewees allocate to prepare the trip or the main motivation that leads them to prepare it. It also identifies skills when planning and booking trips. Another factor that is taken into account to group people is, for example, the number of trips per year, the duration of the trips or the company. The final 6 groups of Braintrust's study are as follows: •“Culturers”: They represent 10% of the market and are those with the highest budget (€ 925 on average). Their motivation is usually to know the heritage, visit the cities to learn their culture or history. They spend a Strain Rtrain Q A door-to-door route optimizer that includes sustainability criteria 22 lot of time thinking and planning the trip. They are usually people who travel alone and the trip is usually long. •“Outsiders”: 7% of the market and those who make the most trips a year (average of 3.2) usually travel with friends or family to learn about the heritage of the place and take the opportunity to visit or spend time with other friends. They tend to spend time planning and go to pre-booked events •“Relax": This group of travelers represents 18% of the total and they allocate a budget below the average. Their motivations are usually national destinations, in search of tranquility and mainly destinations with beaches. It is the group that spends the most time booking the trip and they are usually couples. The average number of trips per year is below the average of the 6 groups and the duration of the trip is usually short. •“Trekkers": Being 18% of the market, this group is usually composed by couples or groups in search of nature and excursions. They tend to make reservations at the destination and spend a budget on the trip below the average •“Hyperconnected": Represent 11% of the market and allocate an average budget of € 857 for travel. This group looks for European destinations and they tend to travel on low-cost airlines. They usually stay in urban hotels or hotels on the beach. They use aggregators and comparators for booking, spending a lot of time and booking well in advance. They tend to be solo travelers. •”Enjoy”: The group that covers the most space in the market, 36%, is usually people who travel accompanied, either as a couple or in small groups whose main motivation is to discover new places and fun activities. They use independent services for reservations and are informed through specialized blogs. Thanks to Braintrust's advanced study, we can divide the traveler market into 6 different groups and assign personalized weights to each of them. The procedure carried out for assigning weights has been as follows: 4 rankings have been made for each of the options (cost, time, footprint and comfort) with the 6 groups explained above. Among the 6 groups, a score between 6 and 1 is distributed successively to each of them to determine the order in which this criterion is perceived as important. (6 points for the group that would give more importance to that criterion when deciding and 1 point for the group that would consider that criterion less relevant for their choice) This classification has been made subjectively based on the information explained above about each group. For example: it has been seen that "Culturers" is the group that allocates the most budget to the trip and their main motivation is to visit specific places to enrich their culture, probably the importance they give to the cost of the trip is less than the rest of the groups, and for that they have received a point (lowest score). On the other hand, the people in the “Relax” group allocate a budget below the 23 average and look for places to relax such as beaches. Perhaps this group would be willing to change the destination of their trip from one place to another if they can find a cheaper offer. This group has received the 6 points (highest score). The rest of points (5, 4, 3 and 2) are assigned to the four remaining groups following the same criteria. The results of this allocation of points according to their relative importance with respect to the rest of the groups are shown in Table 2.5. Table 2.5 Values from 1 to 6 (1 Less important 6 Most important) between groups Once the points have been assigned in the four fields, the next step has been to add individually for each group the total points that have been assigned to them (last column of Table 2.5). The last step has already been to transform the individual score of each section in each group, into a percentage. This has been done by dividing the score for the section by the total sum that that group has of points. In this way, the weight is an independent fraction of the other groups. and within each group, the different criteria receive a percentage of relative importance with respect to the other criteria. For example: the 5 importance points that have been awarded to the group "Hyperconnected" in the time criterion, represent 56% of weight in this group (5/9). On the other hand, the 5 points that the "Relax" group has received for the comfort category, represent only 31% of weight since in other criteria such as cost they have received 6 points, making the sum of total points of this group higher and therefore the fraction that corresponds to the comfort criterion is lower (5/16). Fig 2.1 shows the final weights assigned to each group graphically. Cost Time Footprint Comfort Summation Culturers 1 3 5 6 15 Outsiders 4 6 2 3 15 Relax 6 1 4 5 16 Trekker 5 2 6 2 15 Hyperconnected 2 5 1 1 9 Enjoy 3 4 3 4 14 A door-to-door route optimizer that includes sustainability criteria 24 Fig. 2.1 Weights of different groups of people Culturers 40#% 33#% 20#% 7#% Cost Time Footprint Confort Outsiders 20#% 13#% 40#% 27#% Relax 31#% 25#% 6#% 38#% Trekker 13#% 40#% 13#% 33#% Hyperconnected 11#% 11#% 56#% 22#% Enjoy 29#% 21#% 29#% 21#% 25 As can be seen in Fig 2.1, the weights of the different groups are quite varied. The four criteria are at least once the option with the most weight for some group and the least weight for another. The reason why it has been decided to carry out this procedure to assign user preferences is precisely to be able to experiment with six groups with very different weights in order to check how the solutions vary according to the importance of the different criteria. These weights are fixed and will be used in the advanced study in section 4.3. An experiment will be carried out to determine which is the best transportation system for different routes based on the preferences of different types of user. The following section explains the operation of the software developed to solve the multi-criteria problem of the different means of transport through the VIKOR method as explained in section 2.1. This software also has the different procedures for obtaining data seen in section 1.1 allowing the user to enter only the input data of the trip and know in a few seconds the best transport alternative that suits their preferences. A door-to-door route optimizer that includes sustainability criteria 26 3. SOFTWARE This section explains the procedure carried out to develop a software capable of deciding the optimal means of transport to travel between point A and point B. Next, the operation of the software is exposed step by step from the inputs that the user uses for their search up to the final results provided by the VIKOR method. The complete software works for the console and a web version is currently being developed to have a more attractive interface for the user, but the operation is the same. Step-by-step operation is detailed in Fig 3.1 Fig. 3.1 Steps of optimization software 3.1. Inputs The input phase will be the only moment in which the user will have direct contact with the program and where the optimization process will begin. For this, the user must mainly select the origin and destination locations. This is not limited to the name of a city or town, but in order for the software to be door-todoor, the user will be able to select an exact location thanks to the Google Places API. This API allows the autocompletion of a location while typing its name, the same operation as when using Google Maps. To be able to use it, it must be in the web version since this functionality is integrated into the HTML, in the case of using the console version it will not be possible to fill in with autocomplete but it is possible to search for an exact location by writing the correct way. At this time it will also be necessary to enter the day on which you want to make the trip since the price of the plane tickets will depend on this date. An exact date can be entered as input, the date with the cheapest airline ticket price or the day after the search. Although the prices of train and plane tickets are individual for one person (as is the calculation of the footprint) for the car, it will be necessary to enter the number of passengers for the trip since the cost of the journey will be modified as be divided by the number of passengers, same as the footprint. 27 Finally, for the console version you will need to enter the group of people (seen in section 2.2, "relax", "outsider", etc), to which belongs the person performing the search. In the future web version the user would be automatically assigned to one group and their weights would vary based on the information obtained from the user through questions and search or interest of the user. 3.2. Searches Once the user inputs have been obtained, the program will carry out the corresponding searches to obtain the necessary information. First, the Google Directions API will be used to discover the distance and time that exists by car between the origin and destination location. From the same search, information about the cities and regions to which the locations belong will also be extracted. This information will be useful to locate the nearest airport later. This will help you to locate the nearest airport later. From the API Directions, the train search will also be carried out to also obtain the travel time. For the air route, the procedure consists in first searching the route by car from the origin to the closest airport in the region, and from the airport closest to the destination region to the user's final location. For the flight phase, the SkyScanner API will be used to, first of all, know the IATA code of the airport based on the name of the city. And second, look for the price of the trip between the two airports from its IATA code and the date selected by the user in the inputs. In Annex B it can be seen an example of the response in JSON format to a request to the SkyScanner API about a flight from Barcelona to Madrid. 3.3. Calculations The costs of travel by car and by train are calculated from the respective distances of the route. The same happens with the footprint of the three means of transport as already described in section 1. This phase of the software is where the data obtained in the searches discussed in the previous section are used to calculate the remaining data. For the distance by plane, not yet calculated, the Haversine formula is used through the coordinates of both airports to know the distance in a straight line. 3.4. Matrix Now that all the data needed to compare the different modes of transport are ready, it is time to prepare the data for input into the VIKOR method. To do this, a matrix of 3 rows and 4 columns is built, using the rows for the transport alternatives and the columns for the different criteria of each one.. Table 3.1 is an example of matrix for a search carried out on January 18th 2021, from Castelldefels to Valencia, for February 18th 2021. It is important to note that the cost and footprint data for the car are data for a single passenger, despite the fact that 2 people entered the vehicle in the search. Mij A door-to-door route optimizer that includes sustainability criteria 28 Table 3.1 Matrix with criteria for Castelldefels - Valencia (on Jan 18th to Feb 18th 2021) 3.5. VIKOR This is the step where all the mathematical resolution of the problem happens. The inputs of this step are the data of the matrix of table 3.1; a 1 x 4 vector containing information on whether the criteria are to be maximized or minimized (maximize Comfort and minimize cost, time and footprint) and another 1 x 4 vector with the corresponding weights of the 4 criteria. The output of the VIKOR method is the value that determines the “closest to ideal” of each of the transportation systems. The theoretical explanation of the VIKOR method can be found in section 2.1. 3.6. Outputs Finally, the outputs of the system are achieved, these are the results that the user views on the screen. In order to have a quick visualization of the results, the program will order the rows of the matrix of table 3.1 based on the results of from lowest to highest. In this way, the user will have quick access to price, travel time and footprint data, calculated in the previous steps, with each of the means of transport in order from best to worst solution. It will also be possible to visualize the value of that can allow assessing how close the second and third option are to approaching the best option. Mij Cost (€) Time Footprint (kg CO2) Comfort (%) Car 30,85 3h 12min 37,05 48,8 Airplane 60,00 1h 15min 107,98 5,8 Train 29,99 5h 6min 5,18 45,5 Q Q Q 35 4. RESULTS In this project, the methodology used to calculate (through different models or extracting information directly from third-party services) different data about some means of transport have been explained. These data are: the price, the travel time, the footprint produced by a passenger and the subjective opinion of comfort that can be produced according to the distance of the trip (Section 1.). Thanks to the analysis of decisions with multiple criteria, this information can be combined and turned into a problem to solve which of the different transportation systems can be an optimal solution for a trip, achieving a compromise between the different criteria (Section 2.). In order to solve this problem with different routes between different locations, a software capable of using this methodology has been designed and developed, thus allowing us to know which are the optimal means of transport for making these trips (Section 3.). As it could be seen in section 2.1, the VIKOR method, chosen for the analysis of the routes, allows finding the compromise between all the criteria by adding the option of assigning a relative weight to each one of them. These weights can be distributed in multiple ways. For this project, one of the ways to do it has been through by transforming the preferences of different types of travelers into numerical values to assign different weights to each group (Section 2.2). In this section of the project, the software has been used experimentally to obtain interesting results. Three experiments have been carried out with the same objective, to know the most suitable means of transport to carry out different routes and to analyze the numerical results provided by the software to check how close or how far each of the transportation systems are from being the optimal solution. The first experiment is prepared to answer only two criteria: cost and time. One of the motivations of this project is to be able to bring the user the information on the ecological footprint that can be emitted when planning a trip with one means of transport or another from door to door. However, it is interesting to understand which may be the best solutions taking into account only the two predominant criteria in the usual search engines: economic price and fast trips. In this way, the results of this scenario will serve as a reference with the other two experiments to understand how the best solutions vary once environmental criteria are entered in the searches. For this experiment, the allocation of weights has been 50% for the two criteria, with the intention of finding the best solution that compensates cost and time. In the second experiment carried out, the software configuration was that of a balance between cost, time and footprint. The objective is to check the results of the VIKOR method to the multi-objective problem that is presented only with the respective numerical data of each criterion. That is, without subjective evaluations that determine the weight of one criterion greater than another, or data on comfort. In this way, results can be obtained in which the desire to obtain an economical price and a short journey are equally important, together with the interest in issuing a footprint as little as possible to respect the environment. A door-to-door route optimizer that includes sustainability criteria 36 The third experiment has been prepared to use all the resources of the software developed in this project and to achieve results to be analyzed based on the mission of the project. This experiment therefore has the objective of finding optimal solutions adapting to the profiles of different users, these are the groups of people explained in section 2.2. In this way it will be possible to understand how the application of the VIKOR method varies in this multi-objective problem according to the weights received by each of the criteria; discover, for the routes of the experiment, which are the predominant means of transport as the best solution between the different user profiles and finally compare the relationship that exists between the solutions of this experiment with the scenario in which the criteria are given balanced weights. To be able to compare the results of the three scenarios explained above, it is important that the routes to be analyzed and their cost, time, footprint and comfort values are the same in all three cases. Despite the fact that the software allows to have information in real time such as, for example, the travel time by car and train according to the current situation of roads and train availability, or to get the prices of the plane ticket depending on the day of the search, It has been considered interesting to fix the results of a single search and base the experimental part on the modification of weights or the incorporation and exclusion of some of the criteria. With these experiments it is possible to analyze in a deeper way, how the VIKOR method behaves in the problem of this project, since this methodology is the main nucleus to be able to compare the different means of transport and choose a solution ahead of the rest. Below are the routes chosen to present the results. The location of origin will always be the same: EETAC, Castelldefels which is the place where this project has been carried out. The city of Castelldefels is located south of Barcelona, approximately 11 km from the Josep Tarradelles Barcelona - El Prat Airport and 23 km from Barcelona Estació de Sants, the city's main railway station. •Valencia: 285 km in a straight line. City center •Leganés: 497 km in a straight line. City 30 km from Madrid airport •Lyon (France) 550 km in a straight line. City center •Dos Hermanas: 813 km in a straight line. City 21 km from Seville airport •Paris (France): 843 km in a straight line. City center •Liverpool (UK): 1399 km in a straight line. City center Of the six cities, three of them are national routes and the other three are international routes. On international routes, the energy cost and CO2 emissions emitted according to the railway network used in each country are considered, as already explained in section 1.3.4. Four of the routes have been made to central destinations while Leganés and Dos Hermanas are peripheral cities of Madrid and Seville respectively. Once the routes were chosen, the software was used to proceed with the experiments. As mentioned in section 3.1, the necessary inputs for correct 37 operation are, apart from the origin and destination location, the date on which the trip will be made and the number of passengers. The reason for entering the date is so that the SkyScanner API [SkyScanner API] returns the specific price of that day at the time of the search. Thus, the date on which the experiment is carried out is January 18, 2021. The date selected to make the trip is exactly one month later, February 18, 2021. The reason for selecting the number of passengers is to spread the costs and emissions of the car alternative among the number of people traveling. For plane and train, the data are already calculated per passenger, but for cars this distinction must be made. For this experiment the number of passengers selected is 2. The numerical results of the criteria will be the same for the three experiments since the inputs do not change. These results of cost, time, footprint and comfort are shown in tables 4.1, 4.2, 4.3 and 4.4 respectively. Table 4.1 Search prices by means of transport (€) *Individual search result for two passengers% Table 4.2 Travel times according to the means of transport **Direct route altered by pandemic restrictions Car Price* Airplane Price Train Price Valencia 30,85#€ 60,00#€ 29,99#€ Leganés 52,75#€ 53,59#€ 47,29#€ Lyon 55,97#€ 120,00#€ 52,45#€ Dos Hermanas 80,09#€ 79,90#€ 76,50#€ Paris 85,65#€ 55,65#€ 78,15#€ Liverpool 145,70#€ 134,55#€ 117,95#€ Car Time Airplane Time Train Time Valencia 3h 12min 1h 15min 5h 6min Leganés 5h 54min 1h 31min 4h 2min Lyon 6h 21min 1h 30min 9h 18min** Dos Hermanas 9h 15min 1h 52min 7h 36min Paris 9h 59min 2h 6min 10h 40min Liverpool 18h 24min 2h 45min 22h 24min A door-to-door route optimizer that includes sustainability criteria 38 Table 4.3 Ecological footprint emitted by each of the means of transport (Kg CO2) *Individual search result for two passengers% Table 4.4 Relative comfort of each means of transport based on the distance of the route (%) 4.1. Cost vs time scenario The first scenario of the three is designed to find the best solution based on the two normally predominant criteria when making a trip: cost and time, taking into account a 50% weighing for each of the two. The footprint and comfort criteria are not involved in this experiment. This scenario has the objectives of, first of all, to observe which are the best results according to the distances of the routes and to see if there is any pattern or a specific means of transport that has, in general, a better cost-time relationship than the rest. On the other hand, the results of this experiment will serve as a reference for the other two, and it will be useful to compare once there are results with sustainability criteria or the weights have been modified according to the preferences of the users. Car Footprint* Airplane Footprint Train Footprint Valencia 37,05 107,98 5,18 Leganés 68,98 111,77 8,72 Lyon 73,68 112,85 9,75 Dos Hermanas 108,85 115,89 14,51 Paris 116,95 117,02 14,82 Liverpool 204,52 120 22,68 Car Comfort Airplane Comfort Train Comfort Valencia 48,8 5,8 45,5 Leganés 8,3 37,2 56,3 Lyon 8,0 45,9 46,1 Dos Hermanas 3,8 77,7 17,4 Paris 2,5 95 2,5 Liverpool 0,3 99,4 0,3 39 Next, in Fig 4.1 the results of this experiment are detailed, graphically displaying the resulting value for each of the transportation systems on the six routes that are being evaluated. As explained in section 2.1, the best solutions correspond to the minimization of the value, with 0 being the optimal solution and 1 being the worst solution. Fig. 4.1 values in the six routes from Castelldefels for a “Two-Criteria” scenario The results for this first experiment have been the following: •The car has been the best solution for the routes of Valencia and Lyon (First and third shortest routes of the six) •The train has been the best solution only on the route to Leganés •For the three longest distance routes: Dos Hermanas, Paris and Liverpool, the plane has the best cost-time ratio. A curious result of this experiment is that, on the routes of Valencia, Leganés (near Madrid), Dos Hermanas (near Seville) and Paris, the best solutions have been the same as the options most voted by the public in the survey to determine comfort (Section 1.5). Car to Valencia, train to Madrid and plane to Seville and Paris were the means of transport that respondents considered the most comfortable for these routes. Although comfort does not intervene in this scenario and the only criteria that are taken into account are cost and time, the best solutions have coincided. For the route to Lyon the best option has been the car. Regarding the train, the route provided by the software is quite a long route, with several trains intervening in the process of reaching Lyon. This is due to the current situation Q Q Q Q value in Vikor method 0,00 0,25 0,50 0,75 1,00 Valencia Leganés Lyon Dos Hermanas Paris Liverpool Car Airplane Train A door-to-door route optimizer that includes sustainability criteria 40 that Spain and France are going through and the political measures to contain the coronavirus pandemic. The public transport system at this time is in an exceptional situation, especially if the route that is being sought is between different countries. It is not known if in a normalized global scenario the optimal solution would be different, but as this project was carried out during the coronavirus crisis, the results correspond to a pandemic scenario. It should be noted that the VIKOR method is designed to evaluate problems with multiple criteria and in this scenario only two criteria are being taken into account. This means that perhaps to analyze the relationship between cost and time only of the routes of the experiment, there are better evaluation alternatives to know the best transportation option. However, as explained above, the idea of using the VIKOR method for this scenario is also to be able to compare the current results with those of the following two experiments and assess how the results are modified according to more than two criteria. 4.2. Cost vs time vs emission balanced scenario In this second experiment, the footprint of each of the routes for each transportation system is introduced (Table 4.3). The objective is to find the best solutions to the routes for a balance between the three criteria: cost, time and footprint. All three criteria are equally weighted for the VIKOR method, that is, no one criterion is given more importance over another. In this way, it will be possible to compare with the “two-criteria” experiment how the incorporation of ecological factors affects the resolution of the problem. On the other hand, the results of balanced weights will serve as a reference to the subsequent experiment in which subjectivity plays a fundamental role. The results of Fig 4.2 correspond to the solution of the simulation of the six cases. The value that is represented in the graph is the value of each of the solutions in the VIKOR method. Q Q 41 Fig. 4.2 values in the six routes from Castelldefels for a “Balanced” scenario Of the six cases, the optimal solution for weights of 33% between price, time and footprint, has been different in the different cases, the most optimal means of transport being twice the car, twice the plane and twice the train. The train has never been the worst solution, while the airplane has been twice and the car four times. The routes in which the car has been the best option have been Valencia and Lyon. For the route to Valencia, the airplane penalizes a lot by emitting a much higher footprint than its other two rivals in a short distance as already explained in section 1.2.4. The airplane is the best option for the two longest-distance routes in the experiment: Paris and Liverpool (843 km and 1,399 km respectively). The greater the distance the footprint of the car trip starts to get bigger and closer to the airplane numbers. On the other hand, in these long-distance scenarios neither car nor train can overshadow the much shorter airplane travel time. The train has never been the worst option since the low emissions with a weight of 33% always help the train to have options to be the best choice. However, the two routes that have emerged as the optimal solution have been Leganés (497 km) and Dos Hermanas (813 km), coinciding with the two metropolitan cities of the experiment. It is not surprising that the train is the best option for these two national routes since Spain is the second country with the most high speed lines (3330 km) only behind China [https://uic.org/IMG/pdf/ 20200227_high_speed_lines_in_the_world.pdf]. With the high-speed network, reaching these two cities is fast and at a fairly competitive cost. For this test scenario where price, time and footprint have the same weight, the airplane is the best solution for long and international distances. For mediumdistance routes, the train and the car compete to be the optimal solution while Q Q value in Vikor method 0,00 0,25 0,50 0,75 1,00 Valencia Leganés Lyon Dos Hermanas Paris Liverpool Car Airplane Train A door-to-door route optimizer that includes sustainability criteria 42 the airplane is far from being so. The train network in Spain helps to access parts of the country quickly and cheaply and is always the most ecological solution. For even shorter distances such as 300 km, the car manages to be an optimal solution. If we compare these results with those of the experiment without sustainable criteria, that is, the results of Fig. 4.1 with those of Fig 4.2, it can be seen how the same best solutions are maintained in 5 of the 6 cases. Only on the route to Dos Hermanas does the train become the best solution, replacing the airplane. In any case, in all cases the train happens to have much better values than in the first experiment, being a solution “close to ideality” regardless of the distance of the route. It is, therefore, the mean of transport that most benefits from incorporating sustainability criteria. Car and airplane have worsened their values compared to the first experiment. 4.3. Multi criteria user based scenario This third experiment has the same search inputs as in previous experiments with equal weights, that is, the data of Tables 4.1, 4.2, 4.3 and 4.4. But in addition, it has been added as a fourth input apart from cost, price and footprint, the comfort of each transportation system according to the distance. This experiment is focused on verifying how the results change according to the established weights, that is, trying to know which is the optimal transportation system according to the preferences of the users. For this reason comfort will be the fourth input following the equations that relate it to distance explained in section 1.5. The experiment has been done for each of the six groups of people seen in section 2.2 and using their corresponding weights previously justified. For each group, the optimal travel solution has been sought for each of the six routes: Valencia, Leganés, Lyon, Dos Hermanas, Paris and Liverpool. Figures 4.3 and 4.4 show the sum of times that each mean of transport has been the best and the worst solution respectively for the 6 groups of people. Q Q 43 Fig. 4.3 Number of times that each means of transport has been the best solution among the 6 groups Fig. 4.4 Number of times that each means of transport has been the worst solution among the 6 groups Number of groups 0 1 2 3 4 5 6 Distance (km) 285 497 550 813 843 1399 Car Airplane Train Number of groups 0 1 2 3 4 5 6 Distance (km) 285 497 550 813 843 1399 Car Airplane Train A door-to-door route optimizer that includes sustainability criteria 44 These results show that for short distances the optimal solution between cost, time, footprint and comfort is in most cases the car, while the worst solution is the plane. Quite the opposite occurs for long distances, in which the plane prevails, leaving the car as the worst solution. The train turns out to be the best option in 5 of the 6 cases for the route to Leganés (497 km) and it is the transportation system that less often appears as the worst solution and for at least one of the six groups it is always an optimal solution. Fig. 4.5 shows the average value of that has been obtained for the 6 groups of people on each of the routes. Remember that the best values of are the closest to 0. Fig. 4.5 Average value for the six groups in the six routes from Castelldefels Comparing Fig. 4.5 with Fig. 4.2 from the second experiment we see that: •On the route to Valencia, the same order has been maintained as in a scenario without taking into account weights and comfort, but in this second experiment the value of has improved for train and plane, decreasing from 0.56 to 0.37 and from 1 to 0.81 respectively. •In the case of Leganés, the same thing happens, the same order of means of transport is maintained but the plane improves by reducing its from 0.86 to 0.57 and the car improves significantly, decreasing from 1 to 0.96. Q Q Q Q Q 0,00 0,25 0,50 0,75 1,00 Valencia Leganés Lyon Dos Hermanas Paris Liverpool Car Airplane Train 51 •[Eurocontrol, 2020] Eurocontrol, 2020. All-Causes Delay to Air Transport in Europe November 2020. Retrieved on January 10th 2021 from: https://www.eurocontrol.int/publication/all-causes-delay-air-transporteurope-november-2020 •[Europarl, 2019] European Parliament, 2019. Facts and Figures. Retrieved on February 05th 2021 from: https://www.europarl.europa.eu/ news/es/headlines/society/20190313STO31218/emisiones-de-co2-delos-coches-hechos-y-cifras-infografia •[FlightRadar24] FlightRadar24, 2021. Available at https:// www.flightradar24.com/38.75,-0.89/8 •[Google Directions API] Google Maps Platform. Directions API. Available at: https://cloud.google.com/maps-platform?hl=es •[Hajkowicz and Collins, 2007] Hajkowicz, S.; Collins, K. A review of multiple criteria analysis for water resource planning and management. Water Resour. Manag. 2007, 21, 1553–1566. •[Hwang and Yoon, 1981] Hwang, C. L.; Yoon, K. Multiple attribute decision making: Methods and Applications; 1981. •[ICCT, 2019] Brandon Graver, Ph.D., Kevin Zhang, Dan Rutherford, 2019. ICCT .CO2 emissions from commercial aviation, 2018 •[Ifeu, 2010]. EcoPassenger Environmental Methodology and Data. Retrieved from EcoPassanger: http://ecopassenger.hafas.de/hafas-res/ download/Ecopassenger_Methodology_Report.pd •[Lucidchart] Programa para Hacer Diagramas UML Online. Lucidchart Retrieved on September 2nd 2020 from: https://www.lucidchart.com/ pages/es/ejemplos/diagrama-uml •[NASA, 2011] NASA, 2011. The Polar Jet Stream. Retrieved on December 21th 2020 from: https://svs.gsfc.nasa.gov/3864 •[Pohekar and Ramachandran, 2004] Pohekar SD, Ramachandran M. Application of multi-criteria decision making to sustainable energy planning - Areview.RenewSustainEnergy Rev2004;8:365-81. A door-to-door route optimizer that includes sustainability criteria 52 •[RENFE] RENFE. Available at https://www.renfe.com/es/es •[San Cristobal, 2011] J.R. San Cristóbal. Renewable Energy. Multicriteria decision-making in the selection of a renewable energy project in spain: The Vikor method 36, 2011, 498-502 •[SkyScanner API] RapidAPI. Skyscanner Flight Search API. Available at https://rapidapi.com/skyscanner/api/skyscanner-flight-search? endpoint=5aa1eab3e4b00687d3574279 •[Tallify] All You Need to Know About UML Diagrams: Types and 5+ Examples. Retrieved on February 3rd 2020 from: https://tallyfy.com/umldiagram/ •[UIC] UIC International Union of Railways. Available at https://uic.org •[Via Michelin] Via Michelin. Available at https://www.viamichelin.es 53 ANNEX A door-to-door route optimizer that includes sustainability criteria 54 ANNEX A. Data modeling of the different criteria used to compare means of transport Annex A.1 Model for calculating travel time by car in relation to distance Fig. A.1 Travel time by car against distance according to searches from Castelldefels (Barcelona) in [Via Michelin] for a “Citroën C4 1.6L 110 CV” Annex A.2 Extra cost vs Saved time, compared to cheaper but slower routes Fig. A.2 Relationship between the time gained for choosing the fastest route and the price paid in tolls according to searches from Castelldefels (Barcelona) in [Via Michelin] for a “Citroën C4 1.6L 110 CV” Time (min) 0 200 400 600 800 1000 1200 Distance (km) 300 500 700 900 1100 1300 1500 1700 1900 y = 0,5559x + 62,986 R² = 0,9634 Saved Time (min) 0 17,5 35 52,5 70 Extra Cost (€) 0 35 70 105 140 55 Annex A.3 Price of fuel and tolls against distance according to national routes. Fig. A.3 Price of fuel (blue) and tolls (green) against distance according to searches from Castelldefels-Barcelona in [Via Michelin] for a “Citroën C4 1.6L 110 CV”. Gasoil price: 1.16 €. National routes Annex A.4 Model for calculating footprint by car in relation to distance Fig. A.4 Trend line for footprint vs distance according to searches from Barcelona in [Via Michelin] for a Citroën C4 1.6L 110 CV. Gasoil price: 1.16 € Price (€) 0,00 50,00 100,00 150,00 200,00 Distance (Km) 332 590 593 603 859 917 993 1069 1087 Fuel Cost (€) Toll Cost (€) Footprint (kg CO2) 0 125 250 375 500 Distance (km) 200 425 650 875 1100 1325 1550 1775 2000 y = 0,2161x + 5,3715 R² = 0,9975 A door-to-door route optimizer that includes sustainability criteria 56 Annex A.5 Amount of CO2 proportional to a passenger per km (gCO2 / RPK) Fig. A.5 Share of passenger CO2 emissions and carbon intensity in 2018, by stage length. [ICCT, 2019] 57 Annex A.6 Behavior of the train ticket price according to the search dates Fig. A.6 Average price of the train ticket depending on the date the train leaves according to searches from Barcelona to different destinations in [RENFE] Fig. A.7 Average price of the train ticket depending on the weeks in advance before the train leaves according to searches from Barcelona to different destinations in [RENFE] 30,00#€ 35,00#€ 40,00#€ 45,00#€ 50,00#€ Monday Tuesday Wednesday Thursday Friday Saturday Sunday AVERAGE Weeks before departure 4 5 6 7 Average price (€) 0,00#€ 15,00#€ 30,00#€ 45,00#€ 60,00#€ AVERAGE A door-to-door route optimizer that includes sustainability criteria 58 Annex A.7 Results of the comfort survey Question 1: What means of transport would you consider more comfortable to travel to Valencia? Answers: Question 2: What means of transport would you consider more comfortable to travel to Madrid? Answers: Q2 55#% 37#% 8#% Car Airplane Train Q1 45#% 6#% 49#% Car Airplane Train 59 Question 3: What means of transport would you consider more comfortable to travel to Seville? Answers: Question 4: What means of transport would you consider more comfortable to travel to Paris? Answers: Q3 18#% 79#% 4#% Car Airplane Train Q4 3#% 95#% 3#% Car Airplane Train A door-to-door route optimizer that includes sustainability criteria 60 Question 5: Where would you rather spend a 2 hour trip? Answers: Question 6: Where would you rather spend a 9 hour trip? Answers: Q5 23#% 11#% 66#% Car Airplane Train Q6 36#% 43#% 22#% Car Airplane Train 67 Table C.5 Results of the value for each user profile on the route to Dos Hermanas from the experiment in Section 4.3 Table C.6 Results of the value for each user profile on the route to Paris from the experiment in Section 4.3 Q Car Airplane Train Culturers 0,943 0,01 0,639 Outsiders 1,0 0,01 0,295 Relax 1,0 0,561 0,01 Trekkers 0,948 0,831 0,01 Hyperconnected 1,0 0,01 0,491 Enjoy 1,0 0,01 0,593 SUM 0,9818 0,2387 0,3397 Q Car Airplane Train Culturers 0,959 0,01 0,75 Outsiders 0,941 0,01 0,903 Relax 1,0 0,01 0,511 Trekkers 1,0 0,487 0,096 Hyperconnected 0,95 0,01 0,93 Enjoy 0,976 0,01 0,846 SUM 0,9710 0,0895 0,6727 A door-to-door route optimizer that includes sustainability criteria 68 Table C.7 Results of the value for each user profile on the route to Liverpool from the experiment in Section 4.3 Q Car Airplane Train Culturers 0,998 0,01 0,758 Outsiders 0,83 0,01 0,769 Relax 1,0 0,01 0,278 Trekkers 1,0 0,263 0,01 Hyperconnected 0,867 0,01 0,845 Enjoy 0,997 0,01 0,744 SUM 0,9487 0,0522 0,5673