Transition Towards Fixed-Line Autonomous Bus Transportation Systems

(1)

Transition Towards Fixed-Line Autonomous Bus Transportation Systems

JONAS HATZENBÜHLER

Licentiate Thesis Stockholm, Sweden 2020

(2)

TRITA-ABE-DLT-2011 ISBN 978-91-7873-514-3

KTH Royal Institute of Technology School of Architecture and the Build Environment Department of Civil and Architectural Engineering Division of Transport Planning Brinellvägen 23, SE-100 44 Stockholm SWEDEN Akademisk avhandling som med tillstånd av Kungl Tekniska högskolan framlägges till offentlig granskning för avläggande av licentiatexamen i transportvetenskap Måndag den 25 Maj 2020 klockan 10:00 i K409, Brinellvägen 23, Kungl Tekniska högskolan, Stockholm.

(3)

iii

Abstract

In the last years the steady development of autonomous driving technology has enabled the deployment of more mature autonomous vehicles. These vehicles have been applied in several pilot projects worldwide, most commonly in the form of small buses. At the same time, the amount of people traveling in especially urban areas is continuously growing, resulting in more trips in the transportation system. An efficient transportation system is therefore required to serve the growing passenger demand. Autonomous buses (AB) are assumed to have lower operational costs and with that public transport (PT) systems can potentially be designed more efficiently to facilitate the increased demand better. In this study, an AB specific simulation-based optimization framework is proposed which allows analyzing the impacts AB have on line- based PT systems. The thesis focuses on the transition from existing PT systems towards line-based PT systems operated partially or exclusively by AB.

Existing work on PT service design is extended so that realistic AB systems can be investigated. This is achieved by (i) using AB specific operator cost formulations, (ii) integrating infrastructure costs required for AB operations, (iii) utilizing a dynamic, stochastic and schedule-based passenger assignment model for the simulation of PT networks and by (iv) formulating a multi-objective optimization problem allowing to investigate the stakeholder- specific impacts of AB.

In Paper I the effects of AB, concerning service frequency and vehicle capacity, on fixed-line PT networks are investigated. Among other metrics, the changes are evaluated based on differences in level of service and passenger flow. Additionally, the sequential introduction of AB in existing PT systems is studied. The framework addresses a case study in Kista, Sweden. The study confirmed the initial hypothesis that the deployment of AB leads to an increase in service frequency and a marginal reduction in vehicle capacity.

Furthermore, it could be seen that the deployment of AB increases the passenger load on AB lines and that passengers can shift from other PT modes towards the AB services.

Paper II incorporates a multi-objective heuristic optimization algorithm in the simulation framework. The study investigates changes in transport network design based on the deployment of AB. The differences in user-focused and operator-focused network design are analyzed and the impact of AB on these is quantified. This study is applied to a case study in Barkarby, Sweden where a full-sized, line-based PT network is designed to exclusively operate AB. Among other findings, we show that the autonomous technology reduces the number of served bus stops and reduces the total PT network size. Addi- tionally, average passenger waiting time can be reduced when deploying AB on user-focused PT networks, which in turn leads to a further reduction of user cost.

Keywords: Autonomous Buses, Public Transportation, Network Design, Resource Allocation, Simulation-based multi-objective Optimization

(4)

iv

Sammanfattning

De senaste årens framsteg inom autonom körteknik har lett till mer mog- na autonoma fordon. Dessa fordon har setts tillämpas i flera pilotprojekt över hela världen, oftast i form av små bussar. Samtidigt växer mängden männi- skor som reser, särskilt i stadsområden, kontinuerligt vilket resulterar i fler resor i transportsystemet. Därför krävs ett effektivt transportsystem för att tillgodose det växande antalet passagerare. Autonoma bussar (AB) antas ha lägre driftskostnader och därmed kan system för kollektivtrafik (public transport, PT) potentiellt utformas mer effektivt för att underlätta den ökade efterfrågan bättre. I denna studie föreslås ett AB-specifikt simuleringsbaserat optimeringsramverk som gör det möjligt att analysera effekterna AB har på linjebaserade PT-system. Avhandlingen fokuserar på övergången från befintliga PT-system till linjebaserade PT-system som delvis eller uteslutande drivs av AB.

Befintligt arbete med PT-tjänstdesign utvidgas så att realistiska AB- system kan undersökas. Detta uppnås genom att (i) använda AB-specifika operatörskostnadsformuleringar, (ii) integrera infrastrukturkostnader som krävs för AB-verksamhet, (iii) använda en dynamisk, stokastisk och schemabaserad modell för att tilldela passagerare vid simulering av PT-nät samt genom att (iv) formulera ett multifunktionellt optimeringsproblem som gör det möjligt att undersöka AB: s intressespecifika effekter.

I artikel I undersöks effekterna av AB, med avseende på servicefrekvens och fordonskapacitet, på fasta linjer i PT-nät. Förändringar utvärderas bland annat utifrån skillnader i servicenivå och passagerarflöde. Dessutom studeras den sekventiella introduktionen av AB i befintliga PT-system. Det föreslagna ramverket tillämpas på en fallstudie i Kista, Sverige. Studien bekräftade den initiala hypotesen att utplaceringen av AB leder till en ökning av service- frekvensen och en marginell minskning av fordonens kapacitet. Vidare kunde man se att utplaceringen av AB ökar passagerarbelastningen på AB-linjer och att passagerare kan skifta från andra PT-former mot AB-tjänsterna.

Artikel II integrerar en multifunktionell heuristisk optimeringsalgoritm i ramverket för simuleringen. Studien undersöker förändringar i transportnät- verkets design baserat på implementeringen av AB. Skillnaderna i användarfo- kuserad och operatörsfokuserad nätverksdesign analyseras och AB: s inverkan på dessa kvantifieras. Denna studie tillämpas på en fallstudie i Barkarby, Sve- rige, där ett fullstort linjebaserat PT-nät är utformat för att exklusivt driva AB. Vi visar bland annat att den autonoma tekniken reducerar antalet an- vända busshållplatser och reducerar den totala PT-nätstorleken. Dessutom kan implementeringen av AB på användarfokuserade PT-nät ytterligare för- bättra servicenivån främst genom att minska den genomsnittliga väntetiden per passagerare.

Nyckelord: autonoma bussar, kollektivtrafik, nätverksdesign, resursallo- kering, simuleringsbaserad multifunktionell optimering

(5)

v

Acknowledgements

The help and support of numerous people made my time working on this thesis and conducting the research a very joyful and memorable learning experience. First of all, a big thank you to my main supervisor Erik Jenelius and my assistant supervisor Oded Cats for the open support and guidance during the work on this thesis. I am particularly thankful for your supportive and easy to approach supervision style.

I always appreciated our open discussions and the helpful advice I have received from you.

The research conducted in this thesis was part of the iQMobility project lead by Scania and financed through the Swedish Governmental Agency for Innovation Systems (Vinnova); both of which I would like to thank for the continued input, interest in the research and the cooperation along with the research project.

I would also like to thank the advance reviewer Markus Bohlin for taking the time in assessing my thesis and giving me very valuable comments to further improve the quality of this thesis and the discussion about future research directions.

Special thanks go to all my colleagues at the division. You have managed to transform a simple lunch break into a fun event and supported me with my research when I felt stuck. Thank you all for the laughter and support!

Finally, I would like to express my biggest thanks to my family and girlfriend.

Thank you, mum and dad, for the constant support and open ear. And thank you Emelie, for putting up with my work hours and being the best partner one could ever wish for!

Jonas Hatzenbühler, Stockholm, April 2020

(6)

vi

List of papers

Papers included in the thesis

I. Hatzenbühler, J., Cats, O., Jenelius, E., 2019. Transitioning towards the deployment of line-based autonomous buses: Consequences for service frequency and vehicle capacity. Under review at Transport Research Part A: Policy and Practice, January 2020.

II. Hatzenbühler, J., Cats, O., Jenelius, E., 2019. Fixed-line network design in light of autonomous buses. Submitted to Transportation, February 2020.

(7)

vii

Declaration of contribution

The research motivation and the conceptualization of the research project were done in close discussion with Erik Jenelius and Oded Cats. The generated ideas led to the development of Paper I and Paper II. The research design, implementation, computational experiments, result analysis and writing were done mainly by me.

Both, Erik Jenelius and Oded Cats, helped me with the development and critical assessment of the chosen methodology as well as the interpretation and analysis of the results. They have also taken an active part in the internal revision of the papers.

(8)

List of Figures

3.1 Conceptual overview of Paper I and Paper II. In orange the framework for paper I is shown and in blue the framework for paper II. . . 10

(10)

List of Tables

4.1 Relationship between papers and research objectives . . . 13

x

(11)

List of Acronyms

AB Autonomous Bus

ABC Artificial Bee Colony

AV Autonomous Vehicle

PT Public Transport

RQ Research Question

TNDFSP Transit Network Design and Frequency Setting Problem

(12)

(13)

Chapter 1

Introduction

In recent years the number of pilot projects including driverless vehicles has steadily increased. This is mainly due to advances in technology that enable the vehicles to drive safely in real traffic situations and proactive policymakers who use the pilot projects to study the acceptance among the public. In Ainsalu et al. (2018) the authors present an extensive overview of on-going and completed EU-based projects utilizing autonomous vehicle technology. Most of the ongoing projects and research about the use of autonomous technology have focused on autonomous driving for cars on highways whereas replacing or substituting conventional public transport (PT) systems has been underrepresented. In the last years, however, a growing number of studies dedicated to the synergetic effects of autonomous vehicles (AV) and PT systems have been published (Liang et al., 2016; Scheltes and de Almeida Correia, 2017; Abe, 2019; Tirachini and Antoniou, 2020). These studies investigate concepts like shared autonomous last-mile services, feeder services, and others. The potential autonomous technology has in improving individual traffic systems was shown by (Alonso-Mora et al., 2017). These studies are applied in areas with a large population and a high demand density. The main effects are an increase in people per car and an increase in vehicle-km traveled, which implies that fewer vehicles are required to serve the given demand. In rural areas, similar studies show that a large number of vehicles are required to serve the sparse demand, which in turn means that many vehicles are needed and the number of empty-vehicle kilometers driven is high (Ben-Dor et al., 2019).

To connect high demand urban areas conventional public transport has proven to be an efficient solution. These systems typically operate on a fixed-line network and follow a schedule. The advantage of such operations is the high passenger vol- ume these systems can transport and the simplicity of usage for passengers. One of the main disadvantages of these systems is the low service quality and the inefficient use of resources in low demand areas. Due to autonomous technology, the need for bus drivers is reduced and hence the labor cost reduced. This, in turn, has the

(14)

2 CHAPTER 1. INTRODUCTION

potential to (1) increase the level of service through an increased service frequency, (2) reduce travel times through direct connections, (3) increase operating hours of the bus system and (4) reduce of travel costs, on line-based bus systems. With the introduction of autonomous and electrified driving technologies on line-based new problems arise, e.g. facilitating charging spots, managing vehicle fleets and transitioning towards autonomous bus (AB) operations. Especially the transition scenarios towards AB services, the relation between reduction of operator cost due to autonomous technology and impacts on the level of service has not been studied much in the literature before. We have therefore identified the research gap in exploring the impacts AB has on line-based services.

All of these studies show the potential of autonomous technology when operated in traffic with exclusively autonomous vehicles. There has been a lack of studies that look into the transition towards autonomous induced transportation systems.

Due to practical reasons and regulations, it is not realistic to assume a sudden and fast change from conventional to autonomously driven vehicles in the system, as it is assumed by most current studies. Therefore, a focus area of this work is the investigation of the transition phase towards autonomous vehicles in traffic systems.

This thesis aims to investigate the effect driverless technology has on fixed-line transport systems concerning the level of service and the service design of such systems. The author believes that the fixed-line driverless buses will be the first autonomous systems that will be in long-term operation since they are already in advanced pilot phases, allow for a controlled transition and have lower techni- cal and regulatory burdens. More specifically, this study investigates the effects autonomously operated buses have on the service frequency, vehicle capacity, passenger flow, and network design when operated as a fixed-line service. The changes in resource allocation, sequence of technology introduction and the alignment of the fixed-line network are compared with conventionally operated networks. The presented work focuses on the utilization of small to large-sized buses operating on public roads in mixed-traffic conditions. The knowledge gained through the presented analysis benefits the understanding and impacts of autonomous buses for transport authorities, transport operators, the users of public transport systems and vehicle manufactures.

1.1 Literature Review

The relevant literature for this work is divided into three subcategories. First, studies about the allocation of public transportation (PT) resources in terms of vehicle capacity and service frequency are presented. Then relevant literature concerning the design of transportation networks is highlighted. Lastly, works developing passenger assignment models are briefly discussed. The section closes with the de-

(15)

1.1. LITERATURE REVIEW 3 scription of the identified research contributions.

Resource Allocation

The allocation of vehicle capacity and service frequency on line-based PT networks has been studied over a number of years. A categorization of the previous works can be done based on the method and problem formulation. The majority of the early studies use analytical approaches and single line networks to determine the optimal allocation of vehicle resources. In his study Newell (1971) the optimal headway is determined through a minimization of the passengers’ waiting time.

Salzborn (1972) and Ceder (2007) build on this formulation. They propose formulations integrating trip chaining and variations on demand levels. An important study was presented by Jansson (1980) and Gwilliam et al. (1985). Both authors develop an analytical model to determine optimal vehicle capacities and service frequency, respectively. More specifically, Jansson (1980) derived a mathematical expression considering simplified user costs and operator costs to determine the optimal resource allocation on simple line networks. This "square root formula"

considers user costs and operator costs by integrating passenger flow, maximum crowding level and peak hour duration among other factors. More recent work by Yu et al. (2011) proposes a comprehensive approach for the resource allocation problem. The model considers in-vehicle and bus stop crowding effects, dwell times and denied boardings on PT networks.

The research done on resource allocation of AB on line-based PT networks is limited. Wen et al. (2018) explore the impacts of AB on the modal split of the entire PT system. The authors focus on the last-/first mile use case of AB. The method used is a combination of agent-based simulation and analytical demand model. The authors report that 82% of private car trips would be done with AB instead while 18% of previous bus trips are shifted to AB. In Abe (2019) the potential benefits of AB in Japan are investigated using a survey. The authors report that transit agencies and operators benefit from the automation of line-based systems, while passengers in an urban environment would benefit from flexible autonomous systems. Similarly, Tirachini and Antoniou (2020) use a variation of the square root formula (Jansson, 1980) in their study of the impacts of AB on PT networks. Single line case studies in Germany and Chile are investigated. The authors show that a benefit for operators and users can be expected when operating AB on fixed-line systems.

Simulation models offer an alternative approach to economic assessments of PT systems. These models have the advantage of capturing complex networks more realistically in terms of the relations and passenger flows of multiple PT lines and system dynamics. In Chakroborty (2003) the authors describe the difficulty of modeling resource allocation as a mathematical program due to the discrete nature

(16)

and the complexity (transfers, waiting times, trip chaining). Transit simulation frameworks have therefore been developed to allow efficient assessment of PT networks. In Cats and Glück (2019) an agent-based assignment model (BusMezzo) that captures uncertainties in supply and adaptive user decisions are proposed. In the last years widely used multi-agent simulation tools like (MATSim or SUMO) have been used in studies simulating autonomous vehicle systems. For example, in Leich and Bischoff (2019) the authors compare the level of service provided by fixed-line networks with shared autonomous vehicles (SAV) using MATSim. They report minor improvements in terms of user costs when using SAV, while the costs for operator are increased mainly because of increased vehicle-km. Similarly, Alaz- zawi et al. (2018) use SUMO to study how a AV fleets could be used to reduce traffic congestion. The authors rate different scenarios based on vehicle-km traveled and free-floating vehicle movements.

Network Design

The research branch of public transport network design has been active for many decades. An extensive overview of the conducted research in the fields of transit network design and scheduling is given in Guihaire and Hao (2008a). The authors identify three sub-problems for the network design: A (i) transit network design and frequency setting problem (TNDFSP), (ii) transit network frequency setting and scheduling problem (TNSP) and (iii) transit network design and scheduling problem (TNDSP), respectively. This work focuses on the first sub-problem, i.e.

the TNDFSP.

One of the first works studies of the TNSDSP is done by Lampkin and d. Saal- mans (1967). In their framework, the authors optimize the travel time and vehicle capacity on a single route network. The demand is modeled static and specific for each origin-destination stop pair. The objective function is computed as the sum of user and operator costs, which are represented as monetary values respectively. In more advanced models additional constraints are considered. The authors in Zhao and Zeng (2008); Chien et al. (2003); Chakroborty (2003) consider the number of bus stops on a route, the total number of route transfers and the route directness as constraints of the TNDFSP. With the increased problem complexity advanced optimization algorithms have to be utilized. A variety of heuristic algorithms are implemented in different TNDFSP studies. Fan and Machemehl (2006) present a simulated annealing algorithm. The authors optimize the combined sum of the costs for the user, operator, and unsatisfied demand. The problem constraints are the maximum fleet size, route capacity, maximum number of routes and maximum trip length. In a recent study by Szeto and Jiang (2012) the application of artificial bee colony optimization (ABC) algorithm is discussed. The authors minimize the weighted sum of total transfers and total travel time in the network, while the problem is constrained by the vehicle capacity. In their model, the service frequency is

(17)

1.1. LITERATURE REVIEW 5 determined considering the maximum vehicles per route. In a second work (Szeto and Jiang, 2014) propose a bi-level model for the TNDFSP. In that model, the lower-level problem is the passenger assignment problem which is constrained by vehicle capacity. The upper-level problem is the minimization of passenger transfers. Similar to their previous study the authors use an ABC algorithm. The study PT networks in Winnipeg, Canada, and Tin Shui Wai, Hong Kong.

Passenger Assignment

An essential part of studying the impacts of changes in the PT network is the modeling of passenger decisions. In the past several assignment models have been developed. The models differ can be categorized in frequency-based (Nguyen and Pallottino, 1988; Spiess and Florian, 1989; Cepeda et al., 2006) or schedule-based decision models (Mark D. Hickman and David H. Bernstein, 1997; Tong and Wong, 1998; Poon et al., 2004). In both categories, models have static or dynamic and deterministic or stochastic decision-making processes. Static assignment models assume a constant demand level over a given time period. This prevents these models to capture changes in e.g. waiting time or vehicle crowding over time.

In dynamic models, these time-dependent changes are considered by computing with-in day changes of e.g. travel demand. In deterministic assignment models, a uniform distribution of passenger arrival processes is assumed. The passengers are then assigned to the PT vehicles based on deterministic rules. For short vehicle headways the assumption of uniform arrival distribution has shown to be a good approximation (Jolliffe and Hutchinson, 1975), however if PT vehicles follow a less frequent schedule they arrival processes can not be assumed uniform anymore.

To capture more complex and more realistic arrival patterns stochastic assignment models have been developed. Hence stochastic assignment models capture better uncertainties in the passenger boarding process. In more complex PT networks the stochastic passenger assignment leads to more realistic results, which is the reason why in this study a dynamic, stochastic and schedule-based passenger assignment model (Cats, 2013) is used.

Research Contribution

The contributions of this study are twofold. First, the work on evaluating the impacts of autonomous buses on existing public transport systems in terms of passenger flow and passenger load is building on the existing research on resource allocation on line based PT systems. Additionally, the work on the sequencing of autonomous bus deployment with the developed simulation framework is con- tributing to policy studies about temporal changes in transport planning induces by autonomous buses. Through the study on the sequential introduction of AB into a fixed-line public transport network, it is possible to identify where autonomous

(18)

buses should be deployed first and in what order. The second contribution is ex- tending the existing research in network design and frequency setting with an AB specific simulation-based optimization method, where the simulation uses advanced dynamic passenger assignment models and the optimization algorithm allows for a multi-objective problem formulation.

1.2 Thesis Organization

The remainder of the thesis is organized into four chapters. First, the research objectives of this work are highlighted. Second, the methodology employed to answer the research objectives is presented. Third follows a detailed discussion of the papers and their scientific contributions. The fourth chapter discusses and implications of the key results of the research conducted and gives potential future research directions. The two research papers are included as attachments.

(19)

Chapter 2

Research Objectives

The overarching goal of this thesis is to analyze operational changes in the fixed- lines public transport system when transitioning towards autonomous bus systems.

Consequently, three research questions (RQ) have been identified and are formulated as follows:

RQ1 What are the impacts of AB on line-based PT service design?

A direct consequence of the driver-less operation of buses is the reduction of operation costs of these vehicles. Hence the impacts of reinvesting these operation cost savings optimally in the service considering user and operator costs are analyzed.

The impacts of AB on PT networks are measured in terms of changes in service frequency, vehicle capacity, passenger flow and bus line load based on vehicle technology.

RQ2 In what line sequence should AB be introduced in a line-based PT network?

This RQ is aiming at the introduction sequence of autonomous technology. Due to budget, fleet size or legislative constraints the extension of an autonomously operated fixed-lines service advances in steps. To maximize the expected profit and level of service the sequence of the introduction is crucial to know to transition towards a fully autonomously driven service. The different potential sequences are ranked based on the combined user and operator costs. A distinction is made between path-dependent and path-independent sequences.

(20)

8 CHAPTER 2. RESEARCH OBJECTIVES

RQ3 How should a line-based AB network be designed to best take advantage of the autonomous technology?

Besides the greater flexibility in resource allocation (see RQ1) the reduction of operational costs might also influence the bus route layout and bus stop locations of a public transport network. Due to lower operational costs, longer trips and more frequent stops might be affordable for the operator to do. Since larger areas can be served and the access and egress time to PT access points can be reduced, more passengers might use the PT system. Therefore it is important to study changes in the network design in light of AB and if the network design is sensitive to different stakeholder requirements.

(21)

Chapter 3

Methodology

To assess the impacts of AB in PT networks on passenger flows, vehicle loads and level of service a detailed representation of the network is required. Therefore a simulation-based optimization framework is developed and utilized to evaluate public transport scenarios. The framework is AB specific by (i) considering AB related infrastructure costs (e.g. road enhancements, bus stop layout); (ii) implementing a connectivity constraint, meaning that each AB in the network has to be able to reach every point in the network without human interference and(iii) an AB specific operator cost formulation. To investigate the optimal network design an efficient optimization logarithm is required which allows for large solution spaces.

The methodology used to answer the RQ is based on a combination of mesoscopic simulation software and optimization algorithm. Based on the simulation output the objective values are computed and each solution can be evaluated.

Challenges for the chosen methodology are mainly the impacts of stochasticity on the simulation results and the definition of convergence criteria. To cope with these challenges convergence criteria tests as proposed by (Auger et al., 2012; Bier- laire, 2015) are performed. Stochastic variability in simulation results is reduced by averaging across multiple simulation runs for each solution. The number of required simulations is determined using the concepts from Burghout (2004).

The data needed to perform the simulation and optimization is acquired from multiple data sources. Open-source data from OpenStreetMap (OpenStreetMap contributors, 2017) are used for mapping and spatial representation of a network.

Passenger demand data are based on recorded data (tap-in passenger registrations), demand prediction models (Trafikverket, 2015)and estimates from public transport operators. Other data for vehicle operational cost parameters, optimization parameters, and simulation specific parameters (e.g. value of time, transfer penalty, etc.) are chosen based on recently published literature and widely accepted standards.

In Figure 3.1 the overall framework can be seen. Paper I uses a brute force algorithm in combination with the mesoscopic simulation tool to compute resource

(22)

10 CHAPTER 3. METHODOLOGY

Figure 3.1: Conceptual overview of Paper I and Paper II. In orange the framework for paper I is shown and in blue the framework for paper II.

allocation for a given PT network. A single line analytical model is used to determine the initial solution for the resource allocation problem. The decision variables are service frequency, vehicle capacity, and vehicle technology. The paper also analyses the sequence of autonomous technology introduction. In paper II the PT network design problem is solved using service frequency, bus stop locations and the bus line configuration (number of stops, length of a bus line, the sequence of stops and the routes in-between) as the decision variables. The research problem is solved using a multi-objective ABC and mesoscopic simulation tool.

3.1 Mesoscopic Agent-Based Simulation

With the advantages in computational processing power, a combination of both micro- and macroscopic simulation is possible. Mesoscopic simulation tools allow for the detailed analysis of large, real-world networks and are therefore a helpful tool for traffic prediction, planning, and transport engineering.

For this work the simulation of large transport systems including passenger flow and vehicle movement is important. Additionally, the simulation software has to be capable of simulating public transport systems. This requires the consideration of vehicle schedules, bus routes, concepts for bus stops and the measurement of passenger experiences. OpenSource simulation tools or commercial software pack-

(23)

3.2. HEURISTIC OPTIMIZATION IN NETWORK DESIGN 11 ages tools like SUMO (Krajzewicz et al., 2012), MatSim (Horni et al., 2016) and Vissim (PTV, 2004) all offer the capability to study traffic and transportation system. However, they cannot realistically simulate different passenger assignment processes. Here the open-source simulation tool BusMezzo has the greatest advantage over the other mentioned simulation software and is therefore chosen in this work.

Busmezzo includes a dynamic PT passenger assignment and vehicle simulation model. This model is based on the road network and the PT network definition.

These definitions include road intersections, junctions, lane information as well as information about bus stops, bus lines, bus routes, the level of demand and the vehicle schedule. Passenger boarding, alighting processes and vehicle run and dwell times are stochastic. The stochastic nature might impact the convergence and variance of the objective function, therefore to account for the stochastic processes and minimize their impact on the results each network is simulated multiple times.

The number of simulations is computed using the mean and standard deviation of individual simulation outputs following the process as proposed by Burghout (2004). The decision-making model is based on utility maximization for each passenger. For each passenger, the utility from its origin to destination for multiple paths is computed. During their journey, passengers can reevaluate their paths and make changes based on real-time information on the current traffic situation. The utility is computed based on expected in-vehicle time, waiting times, number of transfers and others. Besides the dynamic passenger assignment, BusMezzo comes with a day-to-day learning option. This option allows for a learning process of the passengers between days. Passengers store the utility of a previously experienced journey and can adapt their behavior according to this utility value. Over the course of multiple days, a passenger can adjust their journey based on these experiences which eventually results in a more realistic and robust network evaluation.

3.2 Heuristic Optimization in Network Design

The transit network design problem as discussed in the second paper is NP-hard (Guihaire and Hao, 2008b; Magnanti and Wong, 1984). To solve the problem in a reasonable time either several simplifications have to be made to the problem formulation or heuristic problem solution methods have to be applied. Past research has considered both approaches: Analytical models to solve the network design problem in a more general way (Ceder and Wilson, 1986; Spiess and Flo- rian, 1989) and optimization models using heuristic algorithms (Chakroborty and Wivedi, 2002; Hadi Baaj and Mahmassani, 1995; Chien et al., 2003; Szeto and Jiang, 2014). Models using Simulated Annealing, Genetic Algorithm, Ant Colony Optimization, NSGA-II, and Artificial Bee Colony Optimization have shown the most promising results in recent years.

(24)

12 CHAPTER 3. METHODOLOGY

With advances in computation and higher availability of transport-related data more realistic and complex transit network models can be developed. In (Zou et al., 2011) the authors have compared different heuristic algorithms and have shown that NSGA-II and ABC are promising algorithms for solving the multi-objective optimization problems. In terms of solution diversity and solution convergence, NSGA-II solutions are dominated by ABC generated solutions. In this study, the ABC algorithm is chosen because of its capability to efficiently examine the solution space and because of the promising results reported in other studies. The ABC requires a relatively large number of evaluations to converge to a solution. For the strategic network design problem as presented in this work the computation time is not crucial and therefore this disadvantage is not relevant.

To formulate the natural behavior of bees into an optimization algorithm the notation of bees and the food source has to be conceptualized. In this work, a bee is represented as one network design including information about the service frequency and vehicle capacity of each line. A food source is the resulting objective values from this network design solution.

The ABC is an iterative algorithm. Each iteration includes the exploration of multiple neighborhoods each by one bee, all of the bees report the objective values to the onlooker bees. After the onlooker bees have received the objective values from all neighborhoods the next generation of employed bees is allocated to the food sources with the best objective values. This mechanism focuses the search towards good quality food sources which helps the algorithm to converge efficiently. For this work, the general ABC, as described above, is adjusted to be able to solve multi- objective problems. To do this the key concepts of exploration and exploitation using three bee types are preserved. Additional heuristics for mutation, ranking of solutions and fitness computation need to be implemented. The proposed adjust- ments to the multi-objective ABC (MOABC) is a consolidation of the works by Zou et al. (2011) and Deb et al. (2002).

3.3 Summary

In summary, the methodology used in this work is a combination of multi-agent PT simulation and heuristic optimization. The first paper uses a brute force approach to evaluate all feasible network scenarios, while the shortest path and myopic algorithms are implemented to extract the recommended sequence of autonomous technology introduction. The second paper utilizes the proposed simulation-based optimization framework where the same simulation software as in paper one is used to evaluate different network designs while the MOABC algorithm allows for an efficient convergence to a solution.

(25)

Chapter 4

Scientific Contributions

This chapter links the research questions to the respective papers included in the thesis (see Table 4). Additionally, the main contributions of each paper are highlighted.

Table 4.1: Relationship between papers and research objectives

Notation Research questions Papers

I II RQ1 What are the impacts of AB on line-based PT

service design? X X

RQ2 In what line sequence should AB be introduced in a line-based PT network? X RQ3 How should a line-based AB network be de-

signed to best take advantage of the autonomous technology?

X

4.1 Paper I

This paper studies the effect autonomous driving technology has on the service frequency and vehicle capacity of fixed-line public transport operations. Additionally, the impacts on passenger flow between modes and consequences for policymakers are extracted. Besides that, a framework to determine the sequence of technology introduction is developed. The main contributions of the paper are addressing the problem of introducing autonomous vehicle technology in a sequential process. This paper applies the methodology to the real-world setting of Kista, connected to the

(26)

14 CHAPTER 4. SCIENTIFIC CONTRIBUTIONS

AB pilot project in that area. This topic has not been discussed in the context of transport service planning before. The sequence of introduction is especially interesting and important for autonomous vehicles since currently, the deployment of actual buses on public roads is a complicated and time-consuming process. It is therefore important to introduce autonomous vehicles on roads and lines which bring the highest advantages to the transportation system.

The results of this study help to further understand the system-level implications of AB and more specifically give a tool to manifest the notion of improved service quality for users of public transport systems through autonomous buses. The developed framework also shows reasonable computational efficiency for a strategic planning tool, witch which realistic (number of routes, level of demand and vehicle fleet) transport scenarios can be investigated. In combination with optimization algorithms as presented in (Paper II) the area of the case study could further be increased. The key findings of paper I are:

• The reduction in operating costs of AB improves the level of service for the PT users. This is primarily due to an increase in service frequency on the AB lines, which translates into shorter waiting times for the passengers. The higher service frequencies also make the lines more attractive for other users and therefore increases the passenger load. Another direct consequence of the AB lines is a slight increase in the recorded transfers (RQ 1).

• The total travel time and the travel time per passenger is not substantially affected. Hence it can be inferred that the deployment of AB on fixed-line services increases the vehicle-km traveled but does not reduce the total time spend by passengers in the PT system (RQ1).

• The best sequence of introduction is different for a user-focused or an operator- focused service design. The user-driven sequence follows two principles. First, deploy AB vehicles on high demand lines. Second, connect the high demand lines with the bus line which has the shortest trip duration. The operator driven sequence follows two simple principles. First, automate the lines with the highest service frequency. Second, automate bus lines with a long cycle time (RQ2).

In summary, this study has confirmed the initial hypothesis, that the reduction of operational costs leads to changed resource allocation for autonomous vehicles.

The service frequency shifts towards higher frequencies, while the vehicle capacity only experiences a minor reduction. AB attracts passengers from other modes like train or metro and therefore increases the load on automated lines. The vehicle-km increases with the deployment of AB.

(27)

4.2. PAPER II 15

4.2 Paper II

Paper II extents the previously presented framework with an optimization model.

It is, therefore, possible to explore larger case studies and investigate more complex research questions. In other words, paper I reveals a shift in service frequency towards higher frequencies and in paper II the changed resource allocation conditions are used to motivate a change in the spatial representation of an autonomously operated PT network. The paper develops a multi-objective simulation-based optimization framework to solve the AB specific network design problem. In addition to the network design aspect, the second paper gives insights into the effects of autonomous technology on the level of service for users of PT systems. In this study, a realistic real-world case study is analyzed. The case study is connected to a pilot project situated in Barkarby, Stockholm. There AB is operated on public roads and integrated into the PT system. In Barkarby new residential areas and metro lines are built in the future. Hence the AB network has to be redesigned to capture these changed conditions. The contributions are summarized in the following list:

• An AB specific transit network design and frequency setting problem (TNDFSP) is proposed, including line connectivity constraints and bus stop position as a decision variable.

• Integration of infrastructure costs and AB specific operator cost formulations in the multi-objective optimization network design problem.

• The network design problem is formulated as a multi-objective problem with user cost, operator cost, and infrastructure costs as the three objectives. The solutions are ranked following distance measures and non-dominated characteristics of each solution.

To investigate the consequences of autonomous technology in terms of length of trips, number of transfers, waiting times and other PT network performance indicators the study is conducted on a real-world test case and benchmark tests.

The key results for this case study and the general findings of the second paper are presented in the following list:

• The deployment of exclusively AB leads to an average reduction of operation cost by approximately 55% compared to conventional buses. Due to the assumed AB specific infrastructure enhancements the costs for infrastructure increase. The reduction in operation cost does not in general result in a reduction in user cost. Even though the service frequency can be increased in most network design solutions in the case of AB deployment (RQ1). Line-based AB services are characterized by shorter network length and fewer bus stops, this leads to an increase in access/egress time and more direct connections (RQ1). In the case study the AB operation led to an increase in transfers and an increase in perceived in-vehicle time.

(28)

16 CHAPTER 4. SCIENTIFIC CONTRIBUTIONS

• High service frequencies, a large network, and many bus stops are indicators of user-focused network design. As a consequence, the network has low waiting times, low access and egress times but high in-vehicle times and a high number of transfers. Operator focused PT networks have only a few lines operated with low frequencies. The network size is small and the bus route only connects the high demand sections (RQ3).

• Introducing AB on user-focused PT networks can promote further improve- ment of the level of service. The main cost driver is the reduction of waiting time through an increased service frequency. The operation of AB on operator-focused design does not result in a reduction in user cost. In the case study, the waiting time could not be reduced through the increase in service frequency, which could be because of an increased demand on the bus line which results in longer boarding processes and crowded bus platforms (RQ2, RQ3). The autonomous technology has, therefore, the biggest impact in terms of user cost on larger networks, in that case, the higher service frequencies lead to network-wide improvements. On single line networks, the assumed operator cost savings cannot translate to a significant user cost reduction.

The main learning from paper II can be divided into two parts. First, it can be said that the operation of AB indeed leads to different network designs. The autonomous technology leads to fewer bus stops and shorter network length. It can also result in an better level of service through the reduction of waiting time.

Second, user-focused network and operator-focused network characteristics in the case of AB deployment are similar to the network characteristics in the conventional case. User-focused design is based on high number of frequencies and number of lines, which gives low waiting times and a many transfers through indirect connections. Operator-focused design is characterized by short and direct bus lines serving the OD pairs with the highest demand. Long waiting times and low total served passengers are the consequences.

(29)

Chapter 5

Implications and Future Directions

In this chapter, the implications of the results for different stakeholders, the methodological limitations, possible extensions, and future research directions are discussed.

5.1 Implications

The stakeholders considered in this work are users of PT services, PT operators, policymakers and manufacturers. For the user, the biggest impact of AB is the increased service frequencies which in turn can lead to a reduction in waiting time.

However, due to the more attractive service, it can be expected that the passenger load on these lines is increased. Additionally, the number of bus stops and network size are slightly reduced compared to conventionally operated networks which leads to slightly longer access and egress times in the case of AB operation. In general user costs can be reduced most efficiently by increasing the number of lines serving the area of interest and increasing the service frequency. This is valid for both, conventional and autonomous bus PT networks. The proposed framework can be used by planners and operators to design AB services. For the operators of AB fleets, the main consequence is the shift in service frequency and the potentially longer operation times. Additionally, the increase in service frequency may induce a reduction in vehicle capacity on shorter, low demand lines. Furthermore, the operation of AB attracts passengers from other PT services, e.g. parallel bus lines.

These network effects may impact the vehicle fleet characteristics, i.e. a change in the number and size of the vehicle fleet required.

Policymakers and authorities should consider the network-wide impacts that high-frequency lines have on the existing public transport network due to the redis- tribution of passengers among the PT lines. In combination with the sequential introduction, path-dependent planning process are preferred over path-independent.

(30)

18 CHAPTER 5. IMPLICATIONS AND FUTURE DIRECTIONS

The consequences of this study for manufacturers of AB are twofold. First, the study is in line with existing research which proposes a trend towards smaller bus capacities, which in turn requires smaller bus vehicles and leads to changed man- ufacturing needs. Second, in areas with high PT demand and especially between high demand OD pairs bigger buses will still be required to facilitate all passenger journeys efficiently. A complete shift towards small buses is therefore not likely.

The increase in passenger load on certain lines and platforms leads to higher capacity requirements for the PT infrastructure. The size of the platforms, stress on the roads and space requirements for potential dedicated lines impact the future urban planning process. A holistic process is therefore advisable in case of success- ful AB integration.

5.2 Discussion

The results in this thesis are subject to some limitations, which are due to the framework design, the methodology chosen and the data sets used. In both presented studies the assumptions for the operator costs of AB are based on reported projects around the world. Due to the limited number amount of published data these assumptions lack robustness. A sensitivity analysis concerning the estimated cost parameters is performed. The result shows that the results are fairly insen- sitive to parameter uncertainty. The used simulation tool incorporates stochastic decision-making processes. The number of replications has therefore been carefully chosen to minimize the risk of statistical outliers. The case studies used to gen- erate the results are located in Stockholm. Consequently, the results cannot be generalized to all other cases without further adjustment of the input and model parameters. The proposed framework is flexible and generally applicable, whereas the detailed results and numerical values are case-specific.

5.3 Future Directions

The proposed framework can be extended with the integration of sustainability considerations. A future transportation system should not only be designed for users and operators but also improve the environmental footprint. The reformula- tion of the optimization problem could be a first start to finding more sustainable PT systems using AB. Another possible direction of research is the combination of freight and passenger transport. The integrated transport of passengers and freight has the potential to reduce the costs and environmental footprint. The presented framework could be extended with a second layer of freight flow to study the sym- biosis effects of these two streams.

(31)

Bibliography

Abe, R., 2019. Introducing autonomous buses and taxis: Quantifying the potential benefits in Japanese transportation systems. Transportation Research Part A:

Policy and Practice 126, 94–113. doi:10.1016/j.tra.2019.06.003.

Ainsalu, J., Arffman, V., Bellone, M., Ellner, M., Haapamäki, T., Haavisto, N., Josefson, E., Ismailogullari, A., Lee, B., Madland, O., Madžulis, R., Müür, J., Mäkinen, S., Nousiainen, V., Pilli-Sihvola, E., Rutanen, E., Sahala, S., Schøn- feldt, B., Smolnicki, P.M., Soe, R.M., Sääski, J., Szymańska, M., Vaskinn, I., Åman, M., 2018. State of the art of automated buses. Sustainability 10. URL:

http://www.mdpi.com/2071-1050/10/9/3118, doi:10.3390/su10093118.

Alazzawi, S., Hummel, M., Kordt, P., Sickenberger, T., Wieseotte, C., Wohak, O., 2018. Simulating the impact of shared, autonomous vehicles on urban mobility – a case study of milan, in: Wiessner, E., Lücken, L., Hilbrich, R., Flötteröd, Y.P., Erdmann, J., Bieker-Walz, L., Behrisch, M. (Eds.), SUMO 2018- Simulating Autonomous and Intermodal Transport Systems, EasyChair. pp. 94–110. URL:

https://easychair.org/publications/paper/28wc, doi:10.29007/2n4h.

Alonso-Mora, J., Samaranayake, S., Wallar, A., Frazzoli, E., Rus, D., 2017. On- demand high-capacity ride-sharing via dynamic trip-vehicle assignment. Pro- ceedings of the National Academy of Sciences of the United States of America 114, 462–467. doi:10.1073/pnas.1611675114.

Auger, A., Bader, J., Brockhoff, D., Zitzler, E., 2012. Hypervolume-based multiobjective optimization: Theoretical foundations and practical implications. Theo- retical Computer Science 425, 75–103. doi:10.1016/j.tcs.2011.03.012.

Ben-Dor, G., Ben-Elia, E., Benenson, I., 2019. Determining an optimal fleet size for a reliable shared automated vehicle ride-sharing service. Procedia Computer Science 151, 878–883. doi:10.1016/j.procs.2019.04.121.

Bierlaire, M., 2015. Simulation and optimization: A short review. Transportation Research Part C: Emerging Technologies 55, 4–13. doi:10.1016/j.trc.2015.01.004.

Burghout, W., 2004. A note on the number of replication runs in stochastic traffic simulation models.

(32)

20 BIBLIOGRAPHY

Cats, O., 2013. Multi-agent transit operations and assignment model. Procedia Computer Science 19, 809–814. doi:10.1016/j.procs.2013.06.107.

Cats, O., Glück, S., 2019. Frequency and vehicle capacity determina- tion using a dynamic transit assignment model. Transportation Research Record: Journal of the Transportation Research Board 14, 036119811882229.

doi:10.1177/0361198118822292.

Ceder, A., 2007. Public transit planning and operation: Theory, modelling and practice. Butterworth-Heinemann, Amsterdam and London.

Ceder, A., Wilson, N.H., 1986. Bus network design. Transportation Research Part B: Methodological 20, 331–344.

Cepeda, M., Cominetti, R., Florian, M., 2006. A frequency-based assignment model for congested transit networks with strict capacity constraints: characterization and computation of equilibria. Transportation Research Part B: Methodological 40, 437–459.

Chakroborty, P., 2003. Genetic algorithms for optimal urban transit network design. Computer-Aided Civil and Infrastructure Engineering 18, 184–200.

doi:10.1111/1467-8667.00309.

Chakroborty, P., Wivedi, T., 2002. Optimal route network design for transit systems using genetic algorithms. Engineering Optimization 34, 83–100.

doi:10.1080/03052150210909.

Chien, S., Dimitrijevic, B., Spasovic, L., 2003. Optimization of bus route planning in urban commuter networks. Journal of Public Transportation 6, 53–79.

doi:10.5038/2375-0901.6.1.4.

Deb, K., Pratap, A., Agarwal, S., Meyarivan, T., 2002. A fast and elitist multiobjective genetic algorithm: Nsga-ii. IEEE Transactions on Evolutionary Computation 6, 182–197. doi:10.1109/4235.996017.

Fan, W., Machemehl, R.B., 2006. Using a simulated annealing algorithm to solve the transit route network design problem. Journal of Transportation Engineering 132, 122–132. doi:10.1061/(ASCE)0733-947X(2006)132:2(122).

Guihaire, V., Hao, J.K., 2008a. Transit network design and scheduling: A global review. Transportation Research Part A: Policy and Practice 42, 1251–1273. URL:

http://www.sciencedirect.com/science/article/pii/S0965856408000888, doi:10.1016/j.tra.2008.03.011.

Guihaire, V., Hao, J.K., 2008b. Transit network design and scheduling: A global review. Transportation Research Part A: Policy and Practice 42, 1251–1273.

doi:10.1016/j.tra.2008.03.011.

(33)

BIBLIOGRAPHY 21 Gwilliam, K.M., Nash, C.A., Mackie, P.J., 1985. Deregulating the bus in- dustry in britain — (b) the case against. Transport Reviews 5, 105–132.

doi:10.1080/01441648508716589.

Hadi Baaj, M., Mahmassani, H.S., 1995. Hybrid route generation heuristic algorithm for the design of transit networks. Transportation Research Part C:

Emerging Technologies 3, 31–50.

Horni, A., Nagel, K., Axhausen, K.W., 2016. The Multi-Agent Transport Simula- tion MATSim. Ubiquity Press.

Jansson, J.O., 1980. A simple bus line model for optimisation of service frequency and bus size. Journal of Transport Economics and Policy 14, 53–80. URL:

http://www.jstor.org/stable/20052563.

Jolliffe, J.K., Hutchinson, T.P., 1975. A behavioural explanation of the association between bus and passenger arrivals at a bus stop. Transportation Science 9, 248–282. doi:10.1287/trsc.9.3.248.

Krajzewicz, D., Erdmann, J., Behrisch, M., Bieker, L., 2012. Recent development and applications of sumo - simulation of urban mobility. International Journal On Advances in Systems and Measurements 5, 128–138.

Lampkin, W., d. Saalmans, P., 1967. The design of routes, service frequencies, and schedules for a municipal bus undertaking: A case study. OR 18, 375.

doi:10.2307/3007688.

Leich, G., Bischoff, J., 2019. Should autonomous shared taxis replace buses? a simulation study. Transportation Research Procedia 41, 450–460.

doi:10.1016/j.trpro.2019.09.076.

Liang, X., Correia, G.H.d.A., van Arem, B., 2016. Optimizing the service area and trip selection of an electric automated taxi system used for the last mile of train trips. Transportation Research Part E: Logistics and Transportation Review 93, 115–129. doi:10.1016/j.tre.2016.05.006.

Magnanti, T.L., Wong, R.T., 1984. Network design and transportation planning:

Models and algorithms. Transportation Science 18, 1–55. doi:10.1287/trsc.18.1.1.

Mark D. Hickman, David H. Bernstein, 1997. Transit service and path choice models in stochastic and time-dependent networks. Transportation Science 31, 129–146. URL: http://www.jstor.org/stable/25768763.

Newell, G.F., 1971. Dispatching policies for a transportation route. Transportation Science 5, 91–105. URL: http://www.jstor.org/stable/25767595.

Nguyen, S., Pallottino, S., 1988. Equilibrium traffic assignment for large scale transit networks. European Journal of Operational Research 37, 176–186.

doi:10.1016/0377-2217(88)90327-X.

(34)

22 BIBLIOGRAPHY

OpenStreetMap contributors, 2017. Planet dump retrieved from https://planet.osm.org.

Poon, M.H., Wong, S.C., Tong, C.O., 2004. A dynamic schedule-based model for congested transit networks. Transportation Research Part B: Methodological 38, 343–368. doi:10.1016/S0191-2615(03)00026-2.

PTV, 2004. User manual, vissim 4.0. Karlsruhe, Germany .

Salzborn, F.J.M., 1972. Optimum bus scheduling. Transportation Science 6, 137–

148. doi:10.1287/trsc.6.2.137.

Scheltes, A., de Almeida Correia, G.H., 2017. Exploring the use of automated vehicles as last mile connection of train trips through an agent-based simulation model: An application to delft, netherlands. International Journal of Transporta- tion Science and Technology 6, 28–41. doi:10.1016/j.ijtst.2017.05.004.

Spiess, H., Florian, M., 1989. Optimal strategies: A new assignment model for transit networks. Transportation Research Part B: Methodological 23, 83–102.

doi:10.1016/0191-2615(89)90034-9.

Szeto, W.Y., Jiang, Y., 2012. Hybrid artificial bee colony algorithm for transit network design. Transportation Research Record: Journal of the Transportation Research Board 2284, 47–56. doi:10.3141/2284-06.

Szeto, W.Y., Jiang, Y., 2014. Transit route and frequency design: Bi-level modeling and hybrid artificial bee colony algorithm approach. Transportation Research Part B: Methodological 67, 235–263. doi:10.1016/j.trb.2014.05.008.

Tirachini, A., Antoniou, C., 2020. The economics of automated public transport:

Effects on operator cost, travel time, fare and subsidy. Economics of Transporta- tion 21, 100151. doi:10.1016/j.ecotra.2019.100151.

Tong, C.O., Wong, S.C., 1998. A stochastic transit assignment model using a dynamic schedule-based network. Trans- portation Research Part B: Methodological 33, 107–121. URL:

https://ideas.repec.org/a/eee/transb/v33y1998i2p107-121.html.

Trafikverket, 2015. Sampers och trafikprognoser - en kort introduktion. Trafikver- ket.

Wen, J., Chen, Y.X., Nassir, N., Zhao, J., 2018. Transit-oriented autonomous vehicle operation with integrated demand-supply interaction. Transportation Research Part C: Emerging Technologies 97, 216–234.

Yu, B., Yang, Z., Sun, X., Yao, B., Zeng, Q., Jeppesen, E., 2011. Parallel genetic algorithm in bus route headway optimization. Applied Soft Computing 11, 5081–

5091. doi:10.1016/j.asoc.2011.05.051.

(35)

BIBLIOGRAPHY 23 Zhao, F., Zeng, X., 2008. Optimization of transit route network, vehicle headways and timetables for large-scale transit networks. European Journal of Operational Research 186, 841–855. doi:10.1016/j.ejor.2007.02.005.

Zou, W., Zhu, Y., Chen, H., Zhang, B., 2011. Solving multiobjective optimization problems using artificial bee colony algorithm. Discrete Dynamics in Nature and Society 2011, 1–37. doi:10.1155/2011/569784.

Transition Towards Fixed-Line Autonomous Bus Transportation Systems