previous day
all days

View: session overviewtalk overview

09:00-10:00 Session 14: Keynote speaker
Location: Kaete Dan
Using game theory to understand public transport in the age of autonomous vehicles
10:00-10:45Coffee Break
10:45-12:45 Session 15A: Passenger data analytics and inferences II

Passenger data analytics and inferences II

Location: Kaete Dan
Share of Strategic Alighting Passengers combining Automatic Passenger Counting and OpenStreeMap
PRESENTER: Rémi Coulaud

ABSTRACT. Understanding passengers’ distribution on-board trains and along public transport platforms is crucial for improving service’s performance and ensure passengers’ comfort. We propose a revealed preference measure of passengers alighting behaviours using automatic passenger counting (APC) data. Our findings revealed that the share of strategic alighting passengers per station is influenced by its layout and the overall passengers volume at this given station.

Inferences of Home Locations using Smartcard Data
PRESENTER: Durba Kundu

ABSTRACT. In order to understand and forecast multimodal travel demand, we need to understand habits and patterns in public transit use. Key locations such as homes or workplaces are valuable for analysing these habits. This research uses transit smartcard (MyWay) data from Canberra, Australia to infer home locations. We present three methods for inferring a home location catchment: an 800-m radius around the most frequently used bus stop for an individual, the Voronoi polygon around the centroid of the strongest cluster of frequently used stops found using K-means clustering, and the convex hull of the strongest cluster of frequently used stops from DBSCAN clustering. We are able to infer a plausible but not validated home location catchment for the majority of smartcard users. 97% of the most frequently used stops fall within 800m of the cluster-predicted home location which have the added benefit of a smaller catchment area. This pilot is a foundation for further work to support transport planning related to home-based trip patterns and home relocation behaviours.

Evaluating route choice-set generation methods for public transport modeling
PRESENTER: Marcela Munizaga

ABSTRACT. One of the biggest challenges when estimating route choice models is to identify the set of routes that are attractive to people when they want to travel between an origin and a destination. This difficulty arises because this set of attractive routes or consideration set is latent, i.e. researcher cannot observe it in a real situation, and therefore requires an explicit path generation process that allows the extraction of the attractive routes. Added to this, the large number of alternative routes that usually are in a transport network, especially in a dense network such as the city of Santiago, make the extraction process even more difficult. The public transport literature shows that researchers have used different techniques to construct the choice set. As Shelat et al. (2019) identified, there are two types of methodologies: (1) direct identification of choice set, and (2) choice set generation methodologies. In the former, usually, the passengers are asked to select an alternative from a knowing choice set for the researcher. As examples of studies that used the direct identification of choice set: Vrtic & Axhausen (2002), which generates different alternatives through varying the value of the attributes, and Eluru et al. (2012), who employed google maps to generate alternatives in the choice set. Some authors have used automatic fare collection technologies in public transport systems to identify directly the choice set (Jánošíková et al., 2014; Kim, I. et al., 2019). In this case, the choice set is conformed by chosen alternatives recorded in the smart card data. Choice set generation methodologies are algorithms that recognize requirements to form the choice set. These methods can be divided into shortest path approaches, constrained enumeration approaches, and probabilistic approaches. The most used approaches are those related to the shortest path, which generates the choice set by iteratively searching the least cost route. Some choice set generation methodologies for the choice set are reported in the literature of public transport route choice modeling. Hoogendoorn-Lanser & Van Nes (2004) applied the branch and bound algorithm (constrained enumeration of candidate paths) for the generation of the choice set in a multimodal network. Guo (2011) worked with the labeling approach proposed by Ben-Akiva et al. (1984). Anderson et al. (2014) adapted the double stochastic method to estimate the choice set in a large-scale multimodal public transport network. Tan et al. (2015) used a combination of four choice set generation approaches: link elimination, labeling, k-shortest path, and simulation. The choice set plays a vital role in the route choice modeling; therefore, it is important to evaluate the different approaches to generate it. However, few studies have investigated the quality of choice sets generated for different methodologies, and all of them are developed with car data. As far as we know, there is no literature on evaluating the choice set generated by different approaches in a multimodal public transport network. Considering that Villalobos and Guevara (2019) found, using synthetic data, that the consideration set constructed with experienced choices of individual performs better than traditional choice set generation methods based on the shortest path, in this work we evaluated the historical choices approach with four traditional deterministic methods: labeling approach, link elimination approach, link penalty approach, and k-shortest path approach. To deal with this purpose we use smart card data from the public transport system in Santiago, Chile. Specifically, we worked with more than 250 origin-destination pairs and data from 20 working days. The explanation of each evaluated approaches is: 1. Historical choices approach: it assumes that the observed trajectories, in a specifically OD pair, conform the choice set considered by passengers. Then, paths are generated by searching observed paths registered in the smart card data for each OD pair. 2. Labelling approach: it assumes that passengers have different objectives. The paths are generated by searching the least-cost paths using different cost functions or labels. We used 7 labels, four of them have only one attribute path in the cost function, and the other three have weight multiple path attributes. 3. Link elimination approach: with this method, the paths are generated iteratively searching the shortest paths by removing the links along the searched path one by one. When all links along the first shortest path have been eliminated, the algorithm move to the next shortest path, and the algorithm stops when the choice set contains 30 paths. The cost that we used contains in-bus travel time, in-metro travel time, waiting time, and walking time. 4. Link penalty approach: with this method, the paths are generated iteratively searching the shortest paths by updating the links cost. When the shortest path is found all their links are penalized by a factor of 1.2. The algorithm stops when the choice set contains 30 paths. The cost that we used contains in-bus travel time, in-metro travel time, waiting time, and walking time. 5. K-shortest path approach: this approach generates the first k shortest loop-less paths from an origin to a destination using the heuristic proposed by Yen (1971). The cost that we used contains in-bus travel time, in-metro travel time, waiting time, and walking time.

We estimated a path-size-logit model with each choice set method and our preliminary results show that the models, calibrated with each choice set approach, capture the effect of travel time in-vehicle, initial and transfer waiting time, and the number of transfers not inside of the metro network. These coefficients obtain negative signs and statistical significance with a confidence level of 95%. Also, in all choice set approaches the path size logit allows capturing the overlap between routes, with a statistically significant positive sign, which is consistent with some evidence reported in the literature of route choice modeling in private transport. However, transfer walking time and number of transfers inside of the metro network are captured with a negative sign and statistical significance with a confidence level at 95% for the historical choices and k-shortest path approach only. Finally, the historical choices approach obtains the best first preference recovery, prediction indicator.

10:45-12:45 Session 15B: Transit service design

Transit service design

Location: Caesarea
Path-Oriented Synchronized Scheduling Using Time-Dependent Data

ABSTRACT. With the emergence of innovations associated with public transport (PT) services, such as Mobility-as-a-Service, demand responsive transit, and autonomous vehicles, the door-to-door PT journey is achievable via multiple transfers between and within different PT travel modes. As such, seamless transfers between different modes of public transportation become an increasingly important factor for the attractiveness of PT services. At the same time, recent developments in travel time prediction methodologies offer new, reliable data sources for the optimization of PT operations. This work, with the consideration of these two elements, develops a mixed integer linear programming model for the PT schedule synchronization problem. The novelty is threefold. First, a novel concept of path-oriented scheduling is proposed. The path transfer time is explicitly formulated and minimized for providing a seamless travel experience considering that the emerging multimodal mobility inevitably induces multiple transfers. Time-dependent travel time data is also utilized in the model, which allows us to harness new and more representative data sources for improving PT services. In order to complement the increase in computational complexity as a result of the utilization of timedependent travel time data, four novel valid inequalities (VIs) are derived. Numerical studies show that the use of time-dependent travel time data is beneficial in terms of reducing path transfer times when compared to using the mean historical travel times and there exists a tradeoff between the maximum allowable path transfer time and the trip time. Meanwhile, simulation studies using Copenhagen network demonstrate that using valid inequalities could significantly reduce the computation time around 20% on average, where the maximum reduction in computation time could exceed 86%. In addition, the proposed valid inequalities are benchmarked against two classes of valid inequalities in the literature. It is found that the proposed valid inequalities could outperform them and combining different valid inequalities could further improve computational performance.

An integer optimization model for allocation of bus lines to the stops of a bus terminal
PRESENTER: Therese Lindberg

ABSTRACT. Interchange stations are essential for a high-quality public transport system. Many passengers pass through a station during the course of a day and the time spent at a station has a large effect on their experience of the whole journey. In this study, we aim to improve the passenger experience at a bus terminal by minimizing the walking distances for all passengers. To this end, an integer linear optimization model which allocates bus lines to the stops of a bus terminal is presented. The model is tested in a numerical experiment using synthetic passenger data. Two alternative approaches, either randomly allocated or based on the number of non-transferring passengers, are used for comparisons. The average improvement in relation to the random allocation strategy is 13%, which shows that the allocation approach has potential. It is thus of interest to collect data from a real bus terminal to further explore the model and the potential benefits it can provide.

The impact of using a naïve approach in the limited-stop bus service design problem
PRESENTER: Homero Larrain

ABSTRACT. Please find attachment

Partially Flexible Demand Responsive Transit services
PRESENTER: Hillel Bar-Gera

ABSTRACT. This study focuses on a partially flexible demand responsive transit (DRT) service for passengers who share one of their trip ends, i.e., either all passengers have the same destination and desired arrival time, or equivalently all passengers have the same origin and departure and time. The main context we consider is a shuttle service from a train station to their final destinations, or nearly equivalently from their origins to the train station. Other possible contexts with similar conditions include airports, schools, industrial parks, community centers, etc. We mathematically formulate the operational challenge of a DRT service of this type as the one-to-many DRT problem (DRT 1-M). The formulation includes vehicle routing as well as assigning passengers to trips and selecting the pickup/drop-off point for each passenger. We assume that the operator guarantees a predetermined level of service in terms of destination-dependent walking time, maximum trip duration, and/or total service time. The operator’s objective function is to minimize vehicle travel time. Preliminary results on modest size problems (Bar-Gera et al., 2019) indicated the potential of this approach. The current study improves the formulations and the solution methods to enable addressing realistic size problems. Based on these new methods, we conduct additional numerical experiments and analyze the results statistically. The solution of the optimization model is done in three stages: 1. Generate an appropriate set of vehicle routes using a dynamic programming algorithm. 2. Select routes to meet the specific demand by solving a set covering problem. 3. Assign passengers to trips. A comparison between alternative dynamic programming algorithms is conducted in terms of efficiency and suitability for several variants of the original formulation. As part of this comparison, we consider the possibility of pre-planning a large set of routes for the entire service area instead of generating the routes only once the demand is known. From a computational point of view, pre-planned routes may allow faster response in real-time. From an operational point of view, pre-planned potential drop-off points may be desirable due to safety considerations. In order to examine how different input characteristics affect the routing solutions of this DRT problem, a sensitivity analysis of the objective function to changes in various parameters of the problem was conducted. The evaluation is based on realistic data regarding train passenger destinations, collected in surveys done by the company ADALYA Economic Consulting LTD for the Israeli Ministry of Transport in different Israeli train stations in 2011-2014. To summarize, the contribution of the current study is in the methodological aspects as well as the operational aspect. The methodological aspects include mathematical formulations of relevant optimization problems, examining solution methods (algorithms and heuristics) and analyzing computational complexity. The operational aspects deal with testing the potential effectiveness and feasibility of this type of service, using a series of numeric experiments and statistical analyses, as well as analyzing the factors that affect the characteristics of the level of service.

10:45-12:45 Session 15C: Passenger demand predictions

Passenger demand predictions

Location: 99 HaYarkon
Effectiveness of trip planner data in predicting short-term bus ridership
PRESENTER: Ziyulong Wang

ABSTRACT. Predictions on public transport ridership are beneficial as they allow for sufficient and cost-efficient deployment of vehicles. At an operational level, this relates to short-term predictions with lead times of less than an hour. Where conventional data sources on ridership, such as Automatic Fare Collection (AFC) data, may have longer lag times, in contrast, trip planner data is often available in (near) real-time. This paper analyzes how such data from a trip planner app can be utilized for short-term bus ridership predictions. This is combined with AFC data (in this case smart card data) to construct a ground-truth on actual ridership. The trip planner data is studied using correlation analysis to select informative variables, that are then used to develop 4 supervised machine learning models (linear, k-nearest neighbors, random forest, and gradient boosting decision tree). The best performing model relies on random forest regression and reduces the error by approximately half compared to a baseline model based on the weekly trend. We show that this model performance is maintained even for prediction lead times up to 30 minutes ahead, and for different periods of the day.

Real-time forecasting of metro origin-destination matrices with high-order weighted dynamic mode decomposition
PRESENTER: Zhanhong Cheng

ABSTRACT. Forecasting the short-term ridership among origin-destination pairs (OD matrix) of a metro system is crucial in real-time metro operation. However, this problem is notoriously difficult due to the high-dimensional, sparse, noisy, and skewed nature of OD matrices. This paper proposes a High-order Weighted Dynamic Mode Decomposition (HW-DMD) model for short-term metro OD matrices forecasting. DMD uses Singular Value Decomposition (SVD) to extract low-rank approximation from OD data, and a low-rank high-order vector autoregression model is established for forecasting. To address a practical issue that metro OD matrices cannot be observed in real-time, we use the boarding demand to replace the unavailable OD matrices. Particularly, we consider the time-evolving feature of metro systems and improve the forecast by exponentially reducing the weights for old data. Moreover, we develop a tailored online update algorithm for HW-DMD to update the model coefficients daily without storing historical data or retraining. Experiments on data from a large-scale metro system show the proposed HW-DMD is robust to the noisy and sparse data and significantly outperforms baseline models in forecasting both OD matrices and boarding flow. The online update algorithm also shows consistent accuracy over a long time when maintaining an HW-DMD model at low costs.

Explaining station demand patterns using Google Popular Times data

ABSTRACT. Google Popular Times (GPT) data are a novel data source that is open to the public, accessible in real-time and available in areas around the world. We aim to explain and predict travel demand patterns for train stations in Kyoto city with this data. Multiple linear regression models are developed using popularity data to analyse the correlation of the station demand patterns and point of interest (POI) visitation rates in the station vicinity. Two of our regression models aim to identify POIs and POI types that have the highest impact on the demand at each station. The other three models use station live popularity, station historical popularity and POI visitation to predict demand. We were able to identify influential POIs and quantify their impacts given that there is a sufficient number of POIs in the vicinity of the station. For prediction, the station’s historical popularity and live popularity are most significant when compared to its surrounding POIs. POI visitation data are shown to be useful when trying to predict further into the future. Our findings suggest that GPT data can enable transit planners and transit users to predict station demand in real-time. City planners would also gain valuable insights in understanding for what type of activities travellers alight or board from each station. Moreover, the method can be scaled and applied to other types of transit stations in other cities.

CB-LSTM Model for Prediction of Crowding in Transit System

ABSTRACT. Please refer to the attached file for the extended abstract.

12:45-14:00Lunch Break (lunch starts at 12:50)
14:00-15:00 Session 16A: Innovative solutions: perceptions and pathways

Innovative solutions: perceptions and pathways

Location: Kaete Dan
Incorporating User Acceptance Probabilities in Optimizing Profit of Carsharing Systems
PRESENTER: Seyma Bekli

ABSTRACT. In this study, we propose a framework for one-way electric carsharing systems which offers spatial/temporal flexible trips to its users. The framework evaluates the user acceptance behavior of the flexible offers and prices, and maximizes the expected profit over possible offer set.

MaaS (Mobility as a Service) Market Futures Explored
PRESENTER: Marcus Enoch

ABSTRACT. The Mobility as a Service (MaaS) concept is powerful because it seeks to rapidly adjust supply and demand for transport so that an (ideally optimal) equilibrium point is attained. Whilst the technological barriers to implementing MaaS are steadily being overcome, less is known about how the MaaS eco-system might evolve.

This paper unpicks the MaaS concept in light of broader societal trends to suggest how it could evolve and offers insights to practitioners and policy makers. The paper draws on literature and discussions with stakeholders to better understand how MaaS has emerged. It constructs four future MaaS market scenarios and identifies implications.

The paper finds current expectations of how the MaaS concept may evolve to be unimaginative and limited in their understanding of how the transport system could change should MaaS be adopted on a wide scale. The major challenges for policy makers will likely relate to balancing the promised benefits offered with issues such as safety (including bio-safety in our post Covid-19 world), data security and privacy, equity and the threat of dominant unscrupulous suppliers distorting the marketplace. Together, these insights suggest that the MaaS reality may be messy and difficult to manage, and that future transport systems might look very different to now.

14:00-15:00 Session 16B: Real-time dispatching and control

Real-time dispatching and control

Location: Caesarea
Principles for setting single or multiline bus holding control based on network characteristics

ABSTRACT. Public transport networks often include one or more sets of common consecutive stops between different lines to serve high-demand segments, or to allow for transfers. In such networks, both single line and multiline control can in principle be applied. In this study, we investigate the effect of both the size of different segments of the network and the characteristics of demand distribution on the performance of single line and multiline control. After introducing the key elements that characterize networks with overlapping segments, two sets of scenarios (a stop set size and a demand-based scenario) are conducted on different network configurations, for both control schemes. Results show that the choice between the two control alternatives is more sensitive to demand distribution than to the lines’ topology. Passenger groups traversing different stop sets are the most consequential in terms of the preferred control strategy. We conclude that multiline shared transit corridor control should be applied for corridors comprising at least 50% of the total number of passengers boarding the common lines.

Dynamic Multiline Vehicle Dispatching Strategy in Transit Operations

ABSTRACT. Providing regularity in buses' operation in high-frequency services is essential to offer a good quality of service to users. If buses are not dispatched at regular headways from the terminal, headway irregularity will gradually increase along the line. In this work, we study a vehicle dispatching problem in which multiple lines start their operations from a common terminal where buses can interchange between lines. The model simultaneously decides the ideal dispatching headway for each line and the line and time in which the following arriving buses to the terminal will be dispatched. The objective is to minimize the dispatching interval’s deviation from an ideal headway that is dynamically updated based on the system’s status. We formulate our problem as a Mixed-integer quadratic problem and adopt a rolling horizon policy to cope with the dynamic and stochastic environment of public transit systems. We prove that a bus assignment that satisfies the FIFO discipline is an optimal solution for the proposed problem. We evaluate our model in a simulation environment under different operational conditions and study the incremental benefits of allowing different flexibility schemes. Our results show that a full flexibility scheme where buses can freely interchange between lines reduces the coefficient of variation of dispatch headways and improves frequency compliance by nearly 20% when compared with the case where buses are restricted to operate in a single line. It also outperforms a myopic heuristic that adopts a priori target headway. The computational times show the feasibility of using the proposed methodology in real-time applications.

14:00-15:00 Session 16C: Real-time railway operations

Real-time railway operations

Location: 99 HaYarkon
Evaluating the influence of train routing selection in real-time railway traffic management
PRESENTER: Marcella Samà

ABSTRACT. To ensure a safe and regular train service, rail infrastructure managers periodically and with large advance design timetables. However, train schedules and operations are subordinated to a large number of parameters, making them vulnerable to service disturbances. Dispatchers are thus required to quickly detect new conflicts arising during operations and take action to recover feasibility, typically by retiming, reordering or rerouting trains in such a way that the propagation of disturbances is minimized. This problem is known in the literature as the real-time Railway Traffic Management Problem (rtRTMP). Several models and algorithms have been developed to solve this problem to provide Decision Support Systems to help dispatcher take more informed decisions [Cacchiani et al. (2014)]. Still, the rtRTMP is an NP-Hard problem and simplifying it is a key aspect to obtain good quality solutions in the short computation time imposed by the real-time nature of the problem.

Some approaches limit the size of the problem by intervening on the granularity used to model the infrastructures and the traffic flows [Pellegrini et al. (2019); Lamorgese and Mannino (2015)], or focus on the solution process to properly drive the search of good solutions [Pellegrini et al. (2015); Samà et al. (2017)]. Others limit the number of variables by considering only what they perceive to be the most significant ones [Van Thielen et al. (2018)]. Often, rerouting variables are the ones most affecting the size of the rtRTMP search space. Thus some work fix the routes in the timetable as the only option, concentrating on the pure scheduling problem [Josyula et al. (2018)], while others use subsets of all the possible alternatives available, which are either based on guidelines set by infrastructure managers [Caimi et. al (2012)], or chosen because considered the ones that will probably lead to the best quality solutions.

The need for a systematic study on the routing selection has lead to the definition of the Train Routing Selection Problem (TRSP) by Samà et al. (2016). The TRSP is the problem of choosing a routing subset for each train so that feasible combinations of train routings exist and good quality solutions for the rtRTMP can be rapidly found. The TRSP is modelled on a construction graph G = (C, L), for which an example is given in Figure 1. Each component ci ∈ C represents an available train routing for a train. Each link lij ∈ L connects two components belonging to different trains if the associated routings satisfy possibly existing rolling stock reutilization constraints between the two trains. Such graph is N-partite and each partition Tt ⊂ C represents all the available alternative routings for a given train t = 1...n. Costs are associated to components and links as an estimation of the influence of choosing a particular routing, on its own or in combination with another one on the final quality of the rtRTMP solution. Solving the TRSP translates into finding a certain number of minimum cost n-vertex cliques, where n indicates the number of total trains considered. We solve the TRSP using an evolution of the algorithm in Samà et al. (2016), based on Ant Colony Optimization (ACO) [Dorigo and Stutzle (2004)], a meta-heuristic inspired by the foraging behaviour of ant colonies. Specifically, the algorithm follows the ACO variant introduced by Solnon and Bridge (2006) for the maximum clique problem. For the TRSP, at each iteration the ants of the colony incrementally build cliques on the construction graph. The probability of selecting a component is computed via the random proportional rule, using pheromone trails and heuristic information. The pheromone trails represent historical information on the quality of the already visited solutions including the component in exam. The heuristic information is a greedy measure of the achievable quality of the solution when such component is added. The validity of this algorithm has been proven when the rtRTMP is microscopically modelled on a track-circuit level with a state-of-the-art mixed-integer linear programming formulation and solved with RECIFE-MILP [Pellegrini et al. (2015)], a decision support tool developed at Université Gustave Eiffel, minimizing the total train delay. This work aims to analyze the impact that solving the TRSP has on the rtRTMP, regardless of the specific model considered for the latter, the objective function optimized and the solution approach used. In addition to the model by Pellegrini et al. (2015) solved with RECIFE-MILP, we consider the rtRTMP microscopically modelled using the alternative graph formulation [Mascis and Pacciarelli (2002)] and solved with AGLibrary [Samà et al. (2017)], a decision support tool developed at Roma Tre University, which minimizes the maximum delay. The two solvers use different strategies for dealing with the rtRTMP. RECIFE-MILP employs a truncated exact algorithm. An initial upper-bound is found for the overall problem by computing a solution for the rtRTMP with fixed timetable routes, before tackling the complete search space with all available routing variables. The solution process stops when either optimality is proven or a maximum computation time is reached. AGLibrary instead uses an iterative meta-heuristic approach for the rtRTMP. Here at each iteration a scheduling solution is computed for the rtRTMP. Then the solution is analyzed to evaluate which rerouting action may transfer train on less congested routes, thus leading to better quality solutions, until a maximum computation time is reached or a solution with no delay is found. A thorough experimental analysis on French railway infrastructures is proposed. Multiple disturbed instances are studied, where disturbances represent train entrance delays in the infrastructure considered. The purpose of the proposed analysis is multi-fold:

– Demonstrate whether the selection of a subset of routing has an advantage not only on RECIFE-MILP but also on the AGLibrary solver;

– Evaluate if changing the rtRTMP objective function needs to be reflected on the TRSP one and/or on the estimation of component and link costs, and how this may be achieved;

– Investigate adjustments to the ACO algorithm regarding the parameters tuning and local search strategies when varying the rtRTMP objective function.

Dependencies in railway operations: Schedule versus reality and resulting prediction difficulties

ABSTRACT. Graph based or equation system based event-driven train delay prediction approaches explicitly model the dependencies of train-events (arrivals, departures) for the prediction horizon. These dependencies arise from the sequence of trains and natural railway operation restrictions. Properly modelling or assuming those dependencies are important in the performance of predictive schemes. As timetables can slightly vary from day to day, their implied dependency structure is not the same for every day. Even more, realised operations do not necessarily follow the same sequence as implied by the timetable. This paper presents a train-event dependency analysis in a busy railway corridor of the Swiss railway network. We thereby provide motivation to build dependency structures dynamically to predict train delays in real-time.

15:00-15:30Coffee Break
15:30-16:30 Session 17A: Equity and fairness in public transport

Equity and fairness in public transport

Location: Kaete Dan
Incorporating Elements of Fairness and Equity in Multiobjective Optimal Spatial Budget Allocation of Transit Services: Application for the City of Nicosia, Cyprus

ABSTRACT. Transit Systems, like cities, can be conceived as living and evolving 'organisms' with ever-changing characteristics. Moreover, transit systems are considered the backbone of today's urban mobility ecosystem and as part of a sustainable mobility paradigm. However, in many cases existing transit systems have not successfully adapted to sociodemographic and technological changes of the urban context. Redesigning a transit system is a painstaking task, and although sustainability concepts aim to incorporate multiple concerns in transport planning (sometimes even misleadingly), transport fairness and equity are often overlooked concepts in the redesign process. This paper proposes a holistic framework for optimally planning a realistic transit system that ensures fairness and transport equity for users. The framework comprises of three main parts: (i) assessing the current sustainability performance of the transit service and infrastructure, equipping a suitable spatial context of tesselated districts (ii) Sustainability, fairness and equity indicators for each district are fitted in structural equations useful for addressing a macroscopic spatial budget allocation purposes, and (iii) Multiobjective PSO optimization for revealing Pareto optimal trade-offs among indicators of sustainability, fairness and equity in transit system development. The study is showcased for the case of Cyprus capital, Nicosia, whose bus public transport system is currently redesigned. The system's essential parts are aggregated at each system level, with administrative areas allowing a macroscopic budget allocation to improve the overall system performance.

15:30-16:30 Session 17B: Public transport data handling

Public transport data handling

Location: Caesarea
The Hubballi-Dharwad Transit ITS, India: Case Study
PRESENTER: Krishna Bajpai

ABSTRACT. This paper describes the case study of ITS implementation in Hubballi-Dharwad for BRT and city bus services.

The 22.5km BRT corridor between the twin-cities of Hubballi and Dharwad provides a true Bus Rapid Transit System in India with all the five components key for being a BRT. In addition, the system also provided a comprehensive ITS system for city-wide bus service.

This case study describes the institutional arrangements, components, the design & implementation framework, challenges and learnings for the HDBRTS ITS project.

Standardizing Transit Survey Data
PRESENTER: Gregory Newmark

ABSTRACT. Transit agencies regularly conduct rider surveys to better understand how customers are using their services. In the United States, on-board surveying became a requirement in 2013 for agencies accepting federal subsidies – an almost universal practice among transit providers. The widespread (and now essentially mandatory) collection of transit passenger data offers an opportunity for researchers to explore patterns across systems and over time. For example, transit planners in Chicago may want to see how their current usage patterns vary from peer systems in New York and Boston or how local usage patterns have changed across waves of surveys. Unfortunately, with limited coordination between survey efforts in different communities and even on the same system across survey efforts, such comparisons typically require a bespoke recoding effort. Such efforts are labor-intensive and therefore costly. Furthermore, since survey instruments do not perfectly align, the recoding entails many assumptions that may hamper comparison. These data comparison challenges can be substantially reduced by standardizing either the initial data collection or the subsequent process of recoding the data collected. This paper examines a sample of recent transit survey efforts and proposes general principles and specific approaches for standardizing transit survey data. Survey instruments were drawn from those collected within the Central Archive of Transit Passenger Data (CATPAD). These instruments were drawn primarily from the busiest transit agencies in the United States as determined by unlinked trips in the National Transit Database. The goal of this work is to facilitate the analysis and therefore enhance the value of transit survey data. Data standardization inherently limits some possible analyses to facilitate others. Guiding principles were developed to serve as a framework for addressing the inherent tradeoffs in designing a standardization policy for transit survey data. These principles include: comparability, policy relevance, consistency, and scalability. The guiding principles were then applied to demographics and trip behavior aspects of transit survey data collection. Ongoing research is investigating standardization of fare policy information. A number of challenges associated with transit data standardization were encountered through this research. These include, among others: • Establishing appropriate thresholds and groupings for ordinal and categorical transit survey variables such as age, income, ethnicity and race • Balancing the competing needs for comparability and flexibility to account for local conditions, such as different languages spoken and different policy • Consistency of location data formats • Ensuring variables remain comparable and relevant over time and space, such as to facilitate cost analysis Based on the framework developed, a proposed approach to transit survey data standardization is presented. The standardization approach seeks to balance data collection considerations with research needs and outlines tiers of standardization that may assist transit agencies and researchers design surveys or conduct analyses that enable comparison across systems and across time, while allowing customization to account for local conditions.

15:30-16:30 Session 17C: Ride-hailing platform operations

Ride-hailing platform operations

Location: 99 HaYarkon
Optimal cancellation penalty for competing ride-sourcing companies under waiting time variability

ABSTRACT. This paper investigates two competing ride-sourcing platforms differing in the service arrival or waiting time variability. It focuses on the impact of order cancellation penalties on each platform’s profit and social welfare. The penalty is modelled by considering a two-stage cancellation behaviour where at the first stage passengers order ride-sourcing vehicles and at the second stage they decide whether to make cancellations. To explore the optimal penalties of different companies, we formulate a game-theoretic framework and obtain the duopoly market equilibrium. This paper uses both homogeneous Poisson and non-homogenous Poisson process to derive the cancellation mechanism, which corresponds to different supply and demand scenarios. We also consider the punishing effect and blocking effect of the order cancellation penalty of the ride-sourcing service. The study fills the gap of incorporating service waiting time variability of two ride-sourcing platforms into order cancellation behaviour, which will produce more effective cancellation penalty schemes for the transportation network companies.

A Generative Model of Ride-Hail Driver Shifts: Time, Duration, and Location
PRESENTER: Gregory Erhardt

ABSTRACT. According to the American Public Transportation Association, by 2019 bus ridership in the United States had declined 14% from its 2012 peak and overall public transport ridership had declined 8% from its 2014 peak. This decline was perceived by the research community as both surprising and alarming, due to its emergence at a period of transit service expansion and economic growth. In 2020, transit ridership had dropped by 79% compared to 2019 levels at the start of the COVID-19 pandemic. While some riders have returned to use transit, ridership in the second half of 2020 and first quarter of 2021 remained 65% below pre-pandemic levels. Continuous efforts are being made to account for these negative trends, yet it seems that a more comprehensive understanding is needed, considering not only users’ perception and acceptance of transit but also those of competing modes, including that of Transportation Network Companies (TNC) that offer faster door-to-door services. TNC, or ride-hailing services, have been identified as a key contributor to declining transit ridership. Changing the current public transportation outlook of companies/drivers operating independently on fixed routes to maximize their own profit, with an outlook that offers riders better and faster services, the idea of a coordinated on-demand multi-modal transit was put forward. Such a service would include the centralized dispatching and routing of micro-transit vehicles, potentially increasing ridership and, in turn, reducing congestion. The present work is part of a project aiming to develop a multi-modal optimization model with a multi-agent simulation considering both transit and TNC as a major competitive mode. Towards this end, we aim to obtain a realistic representation of ride-hailing driver behavior for generating a ride-hailing driver fleet that will be embedded in the simulation alongside transit supply and overall demand. To date, research on ride-hailing services has been hampered by a lack of data. However, we take on a data-centric approach, leveraging a unique data set of ride-hail vehicle traces scraped from the Application Programming Interfaces (APIs) of two ride-hail companies (Uber and Lyft) operating in San Francisco, CA., during a 31-day period between November-December 2016. As the specific Uber data does not include unique driver identifiers, we used Lyft data to generate the ride-hailing fleet for a typical weekday. A total of 1,344,319 trips and 117,579 shifts were extracted from this data set for San Francisco’s 981 Traffic Analysis Zones (TAZs). Using discrete choice models, we modeled ride-hailing driver behavior throughout a 24-hour simulation day within four steps, modeling: (1) number of shifts on the simulation day, (2) shift duration, (3) shift starting time, and (4) shift starting location. Given that drivers often take short breaks by disconnecting from the TNC app, we defined a shift as a consecutive working time that does not include breaks longer than one hour. Given a lack of access to drivers’ socio-demographic data, and trying to account for some of the heterogeneity observed between different drivers, we constructed a driver type variable to differentiate between fulltime drivers (working 35+ hours per week), part-time drivers (5-35 hours) and occasional drivers (less than 5 hours). In step (1), with two dummy variables corresponding to fulltime and occasional drivers, we estimated a model to predict drivers’ number of shifts on the simulation day with choice alternatives corresponding to 0, 1, 2, or 3+ shifts. This model demonstrated that fulltime drivers are more likely to work more shifts per day, while occasional drivers are less likely to do so. For steps (2) and (3), separate models were estimated for drivers’ primary, secondary, and tertiary shifts on the working day. These were defined based on shift length, such that the longest shift on the working day was identified as the primary shift, the second longest was identified as secondary, and all other shifts as tertiary. Fulltime and occasional drivers dummy variables significantly contributed to the estimation of drivers’ shift duration in step (2), yielding positive coefficients for the former and negative coefficients for the latter. A number of shifts variable was also included, suggesting that drivers with more shifts are likely to work longer shifts (thereby also capturing some of the variability related to fulltime drivers). Estimating drivers’ shift starting times in step (3), yielded significant coefficients for driver type, number of shifts and shift duration variables. As can be expected, results revealed two peak periods for shifts’ starting times during the simulation day, one around 8AM and another around 6PM. For step (4), separate models were estimated for the first, second and third shift of the working day, corresponding to shifts’ chronological order, for considering the relation between the end location of drivers’ previous shift and start location of next shift. To estimate the starting location models, we employed a choice modeling approach to first distribute the shifts between San Francisco’s 14 districts. Subsequently, we distributed the shifts within districts proportionally to the population of each of the 981 TAZs. Three main components were incorporated in the utility functions of the choice models: (a) a size term, for measuring the quantity of opportunities, including coefficients for population and employment counts, (b) a utility term, for qualitative measures related to district characteristics, including population and employment, income, and college enrolment densities, and (c) for subsequent shifts, distance from previous shift end location and shared district border dummy. As expected, results indicated a substantially higher number of shifts starting in the downtown TAZs, where population and employment counts and densities are higher. Further, more shifts were found to start in TAZs of higher-, as opposed to lower-, income, and less shifts were found to start in TAZs of higher college students density. Based on the results of these models, a ride-hailing driver fleet was generated for a typical weekday in San Francisco, to be embedded in a multi-agent simulation of the transit system. Ultimately, this simulation will provide transit agencies with a practical tool allowing them to be nimbler in adapting to dynamic conditions. Moreover, this simulation will potentially support planning of coordinated on-demand multi-modal transit.

16:30-17:00 Closing session

Closing session

Location: Kaete Dan