O. Cats
Please Note
36 records found
1
When Schedules Meet the Streets
Public transport reliability, local traffic exposure, and mitigating measures in urban bus and tram segments
A multidimensional segment-hour framework is developed and applied to 15 selected directed bus and tram segments in Geneva, Switzerland. GTFS Schedule data, GTFS Realtime data, and hourly road traffic counter data are combined into a dataset of 228,094 valid segment-hour observations covering 2019-2024. Reliability is represented through three complementary indicators: OTP for schedule adherence, CVTT for travel time variability, and CVh for headway regularity. Local traffic exposure is operationalised using an intensity/capacity ratio. The empirical analysis combines descriptive and within-segment analyses with a hypothesis-guided blockwise OLS regression strategy and outcome-specific robustness checks.
The clearest and most consistent relationship is found for travel time variability. Higher local traffic exposure is associated with higher CVTT in both the OLS and log-transformed specifications, indicating less stable realised stop-to-stop travel times under more intensive traffic conditions. For OTP, local traffic exposure is positively associated with the outcome after controlling for the broader operating context, contrary to the hypothesised direction, while downstream punctuality is strongly associated with upstream delay. For CVh, the traffic relationship is weak and not robust to log transformation.
Dedicated public transport infrastructure and traffic signal priority are represented as a mitigating-measure package. The package is not consistently associated with improved reliability. However, for CVTT, the interaction between local traffic exposure and the mitigating-measure package is negative and weakly significant across both model specifications, providing limited evidence that travel time variability is less sensitive to increasing traffic exposure when the package is present. These relationships are observational and should not be interpreted as causal effects.
The study demonstrates that public transport reliability should be treated as multidimensional and that indicator choice strongly affects the conclusions drawn about traffic exposure and mitigating measures. The framework provides a transferable approach for integrating planned operations, realised operations, and local traffic data in detailed segment-level reliability analysis. ...
A multidimensional segment-hour framework is developed and applied to 15 selected directed bus and tram segments in Geneva, Switzerland. GTFS Schedule data, GTFS Realtime data, and hourly road traffic counter data are combined into a dataset of 228,094 valid segment-hour observations covering 2019-2024. Reliability is represented through three complementary indicators: OTP for schedule adherence, CVTT for travel time variability, and CVh for headway regularity. Local traffic exposure is operationalised using an intensity/capacity ratio. The empirical analysis combines descriptive and within-segment analyses with a hypothesis-guided blockwise OLS regression strategy and outcome-specific robustness checks.
The clearest and most consistent relationship is found for travel time variability. Higher local traffic exposure is associated with higher CVTT in both the OLS and log-transformed specifications, indicating less stable realised stop-to-stop travel times under more intensive traffic conditions. For OTP, local traffic exposure is positively associated with the outcome after controlling for the broader operating context, contrary to the hypothesised direction, while downstream punctuality is strongly associated with upstream delay. For CVh, the traffic relationship is weak and not robust to log transformation.
Dedicated public transport infrastructure and traffic signal priority are represented as a mitigating-measure package. The package is not consistently associated with improved reliability. However, for CVTT, the interaction between local traffic exposure and the mitigating-measure package is negative and weakly significant across both model specifications, providing limited evidence that travel time variability is less sensitive to increasing traffic exposure when the package is present. These relationships are observational and should not be interpreted as causal effects.
The study demonstrates that public transport reliability should be treated as multidimensional and that indicator choice strongly affects the conclusions drawn about traffic exposure and mitigating measures. The framework provides a transferable approach for integrating planned operations, realised operations, and local traffic data in detailed segment-level reliability analysis.
Assessing Robustness of Long-Distance Travel Accessibility
A comparative multi-layer network robustness analysis on the timetable-based aviation and high-speed rail networks in Europe and China}
To answer this question two complementary multilayer transport network representations are constructed. The structural representation keeps rail and air as coupled but distinct layers whose edges are the segments between consecutive stations, the non-stop flights, and the public-transport links among the terminals of the same urban area, while the functional representation uses the structural network as its topological layer and extends it with the scheduled services that realise each connection, carrying their operator, timetable, and generalised travel cost. The two are deliberately kept apart rather than merged into one graph, so that a disruption can be aimed at the structural or the supply level independently while the shared node set keeps the two sets of results comparable. On the functional network a path-enumeration algorithm constructs the 482,947,150 feasible itineraries between 1,674 rail stations and 197 airports in 30 countries, and robustness is measured mainly through the retained efficiency R(x) and the share of baseline itinerary quality that survives a disruption. Three experiment families are evaluated. Structural stress tests remove edges by unweighted topological betweenness, by a service-aware betweenness set by the itineraries each edge carries, or at random. Operator withdrawals and cooperation-tier restrictions act on the 82 rail and air operators. Two documented macro-shocks, the Iceland 2010 volcanic eruption and the July 2021 Bernd flood, test the network's response to real hazards.
Incorporating service-level information does not merely lower the measured robustness, it changes which edges are deemed critical. The two rankings share not a single rail edge among their top fifty, since topology elevates the international cut-edges that keep the continental graph connected while service load concentrates 88 per cent of the top two hundred edges on the German Rhine-Ruhr and Frankfurt-Mannheim corridors. Removing the top five per cent of edges leaves retained efficiency at R(x) = 0.168 in service-aware order against 0.557 in topology order, and against the random benchmark the topology failure is statistically indistinguishable (z_R = -0.47) while the service-aware failure is an extreme outlier (z_R = -8.51). The damage falls as outright disconnection rather than delay, so the service layer collapses long before the largest connected component does. Rail-air co-existence does not offset this concentration. The robustness of the hub-and-spoke air layer does not transfer to the combined network, since an air leg reaches it through rail-side airport-feeder edges whose removal severs the intermodal itinerary at its rail end. The two macro-shocks test this in both directions. An ash closure grounding 77.7 per cent of active airports still leaves R(x) = 0.875 through the immunity of the rail-only baselines, with not a single air baseline replaced by a within-tolerance rail substitute. The Bernd flood lowers intra-German retained efficiency to R(x) = 0.510, with approximately 81,000 structurally feasible air substitutes arriving outside the reroute tolerance. In neither direction does cross-modal rescue materialise.
In sum, robustness is not a fixed property of the European air and rail networks. It depends on how the network is represented, on how severe and how targeted the disruption is, and on how service is organised across operators, from the concentration of supply in a few of them to the cooperation rules that decide whose services a passenger may combine. With service-level characteristics incorporated, they prove substantially less robust to disruptions than their topology suggests. ...
To answer this question two complementary multilayer transport network representations are constructed. The structural representation keeps rail and air as coupled but distinct layers whose edges are the segments between consecutive stations, the non-stop flights, and the public-transport links among the terminals of the same urban area, while the functional representation uses the structural network as its topological layer and extends it with the scheduled services that realise each connection, carrying their operator, timetable, and generalised travel cost. The two are deliberately kept apart rather than merged into one graph, so that a disruption can be aimed at the structural or the supply level independently while the shared node set keeps the two sets of results comparable. On the functional network a path-enumeration algorithm constructs the 482,947,150 feasible itineraries between 1,674 rail stations and 197 airports in 30 countries, and robustness is measured mainly through the retained efficiency R(x) and the share of baseline itinerary quality that survives a disruption. Three experiment families are evaluated. Structural stress tests remove edges by unweighted topological betweenness, by a service-aware betweenness set by the itineraries each edge carries, or at random. Operator withdrawals and cooperation-tier restrictions act on the 82 rail and air operators. Two documented macro-shocks, the Iceland 2010 volcanic eruption and the July 2021 Bernd flood, test the network's response to real hazards.
Incorporating service-level information does not merely lower the measured robustness, it changes which edges are deemed critical. The two rankings share not a single rail edge among their top fifty, since topology elevates the international cut-edges that keep the continental graph connected while service load concentrates 88 per cent of the top two hundred edges on the German Rhine-Ruhr and Frankfurt-Mannheim corridors. Removing the top five per cent of edges leaves retained efficiency at R(x) = 0.168 in service-aware order against 0.557 in topology order, and against the random benchmark the topology failure is statistically indistinguishable (z_R = -0.47) while the service-aware failure is an extreme outlier (z_R = -8.51). The damage falls as outright disconnection rather than delay, so the service layer collapses long before the largest connected component does. Rail-air co-existence does not offset this concentration. The robustness of the hub-and-spoke air layer does not transfer to the combined network, since an air leg reaches it through rail-side airport-feeder edges whose removal severs the intermodal itinerary at its rail end. The two macro-shocks test this in both directions. An ash closure grounding 77.7 per cent of active airports still leaves R(x) = 0.875 through the immunity of the rail-only baselines, with not a single air baseline replaced by a within-tolerance rail substitute. The Bernd flood lowers intra-German retained efficiency to R(x) = 0.510, with approximately 81,000 structurally feasible air substitutes arriving outside the reroute tolerance. In neither direction does cross-modal rescue materialise.
In sum, robustness is not a fixed property of the European air and rail networks. It depends on how the network is represented, on how severe and how targeted the disruption is, and on how service is organised across operators, from the concentration of supply in a few of them to the cooperation rules that decide whose services a passenger may combine. With service-level characteristics incorporated, they prove substantially less robust to disruptions than their topology suggests.
To this end, this study develops a comprehensive simulation framework that combines an advanced ride-pooling candidate generation algorithm (ExMAS) with an innovative two-stage vehicle assignment optimization process, and integrates a nested Logit model to quantify the competition structure with public transport. This framework is applied to 37 Dutch cities of varying scales to derive generalizable findings.
The study finds that population scale is a fundamental determinant of the required fleet size, exhibiting a strong linear relationship, especially in large cities. The key to enhancing operational efficiency, however, lies in more nuanced urban structure metrics. In large cities, longer commuting distances combined with more complex networks (higher node degrees) foster ride-pooling potential. In contrast, for small cities, the local compactness of the network (higher clustering coefficient) and demand density are critical for improving vehicle turnover efficiency. Furthermore, the research confirms that a uniform pricing strategies are unlikely to achieve best performance on pooling efficiency, highlighting the necessity of implementing differentiated pricing based on user preferences and city scale. In competition with public transport, the study's Public Transport Competitiveness Index reveals a non-monotonic relationship with travel distance. A state of competitive balance is observed for short-distance trips, while PT holds a distinct advantage in the medium-distance range. For long-distance trips, the inherent speed advantage of SAVs allows them to become the more competitive option. This external competition also structurally increases the internal pooling rate of the SAV system.
The findings of this study provide critical strategic insights for SAV operators and policymakers, emphasizing the critical importance of adjusting fleet management, service design, and pricing strategies according to specific urban characteristics and the competitive environment, thereby providing decision support for achieving an efficient and sustainable urban transportation system. ...
To this end, this study develops a comprehensive simulation framework that combines an advanced ride-pooling candidate generation algorithm (ExMAS) with an innovative two-stage vehicle assignment optimization process, and integrates a nested Logit model to quantify the competition structure with public transport. This framework is applied to 37 Dutch cities of varying scales to derive generalizable findings.
The study finds that population scale is a fundamental determinant of the required fleet size, exhibiting a strong linear relationship, especially in large cities. The key to enhancing operational efficiency, however, lies in more nuanced urban structure metrics. In large cities, longer commuting distances combined with more complex networks (higher node degrees) foster ride-pooling potential. In contrast, for small cities, the local compactness of the network (higher clustering coefficient) and demand density are critical for improving vehicle turnover efficiency. Furthermore, the research confirms that a uniform pricing strategies are unlikely to achieve best performance on pooling efficiency, highlighting the necessity of implementing differentiated pricing based on user preferences and city scale. In competition with public transport, the study's Public Transport Competitiveness Index reveals a non-monotonic relationship with travel distance. A state of competitive balance is observed for short-distance trips, while PT holds a distinct advantage in the medium-distance range. For long-distance trips, the inherent speed advantage of SAVs allows them to become the more competitive option. This external competition also structurally increases the internal pooling rate of the SAV system.
The findings of this study provide critical strategic insights for SAV operators and policymakers, emphasizing the critical importance of adjusting fleet management, service design, and pricing strategies according to specific urban characteristics and the competitive environment, thereby providing decision support for achieving an efficient and sustainable urban transportation system.
Analyzing route choice preferences in response to planned public transport disruptions
A study of the Washington DC metro smart card data
Descriptive analysis reveals that although travel times and wait times remained relatively stable, passengers shifted from preferring direct routes to favoring routes with more transfers but shorter in-vehicle times after the disruption. Discrete choice models, including Multinomial Logit and Mixed Logit models, were estimated to explore these shifts, but the results showed unexpected coefficient signs, likely due to strong multicollinearity between travel time and transfers, and the presence of dominated alternatives. While the Mixed Logit model improved the model fit slightly, practical interpretability remained limited.
The findings suggest that disruptions can lead to lasting behavioral changes, with passengers reassessing their travel options rather than returning to previous habits. Although AFC data is valuable for detecting such shifts, it alone is insufficient to fully capture the factors driving route choice behavior, highlighting the need for complementary perception-based data in future research. From a policy perspective, understanding these behavioral adaptations can help transit agencies design better service recovery strategies that sustain ridership after disruptions, supporting broader goals of promoting sustainable urban mobility. ...
Descriptive analysis reveals that although travel times and wait times remained relatively stable, passengers shifted from preferring direct routes to favoring routes with more transfers but shorter in-vehicle times after the disruption. Discrete choice models, including Multinomial Logit and Mixed Logit models, were estimated to explore these shifts, but the results showed unexpected coefficient signs, likely due to strong multicollinearity between travel time and transfers, and the presence of dominated alternatives. While the Mixed Logit model improved the model fit slightly, practical interpretability remained limited.
The findings suggest that disruptions can lead to lasting behavioral changes, with passengers reassessing their travel options rather than returning to previous habits. Although AFC data is valuable for detecting such shifts, it alone is insufficient to fully capture the factors driving route choice behavior, highlighting the need for complementary perception-based data in future research. From a policy perspective, understanding these behavioral adaptations can help transit agencies design better service recovery strategies that sustain ridership after disruptions, supporting broader goals of promoting sustainable urban mobility.
B&B is effective only on small scenarios (small Κ and/or modest network size) and becomes intractable as either grows. Greedy with Ν=2 improves over Ν=1, while deeper horizons Ν>2 offer limited or sometimes negative returns. Reverse greedy typically surpasses forward under severe damage. GRASP reliably improves on greedy within minute-level budgets, with swap gains plateauing for small disruptions. Overall, short-horizon reverse greedy, complemented by lightweight GRASP, provides effective and timely recovery plans; B\&B is best reserved for benchmarking and certification on small cases. ...
B&B is effective only on small scenarios (small Κ and/or modest network size) and becomes intractable as either grows. Greedy with Ν=2 improves over Ν=1, while deeper horizons Ν>2 offer limited or sometimes negative returns. Reverse greedy typically surpasses forward under severe damage. GRASP reliably improves on greedy within minute-level budgets, with swap gains plateauing for small disruptions. Overall, short-horizon reverse greedy, complemented by lightweight GRASP, provides effective and timely recovery plans; B\&B is best reserved for benchmarking and certification on small cases.
A Model-Based Characterisation of Delay Propagation in Metro Networks
Developing a Data-Driven Framework for Discovering and Quantifying Delay Behaviour
The proposed framework reduces the modelling of delay propagation to the statistical fitting of relationship functions between network elements using subsets of available variable data. The methodology follows five general steps: defining relationship structures, setting analysis dimensions and their granularity, selecting the method for fitting the relationship functions, fitting the functions using available data, and quantifying the residual variability. This framework is applied to the Washington D.C. Metro using schedule and operational train movement data to analyse and characterise delay propagation behaviour.
Two model implementations are developed. The first explores the full breadth of variables available in the delay data to identify those offering statistically significant delay relationships, revealing that propagation occurs predominantly between directly connected stations and along shared lines. The second, more targeted model quantifies these relationships, demonstrating that propagation strength is independent of delay magnitude and mostly consistent across different time periods. Further cross-examination revealed minimal sensitivity to operational and network design variables, such as headways and connectivity, suggesting that other variables like scheduling regimes might be the cause of the different observed delay propagation behaviours.
The results highlight that localised delay effects dominate over network-wide influences, suggesting that metro operators can improve service reliability by focusing mitigation efforts on key inter-station connections rather than entire lines. The framework offers a versatile and multiplicative foundation for future applications in delay propagation prediction and analysis. ...
The proposed framework reduces the modelling of delay propagation to the statistical fitting of relationship functions between network elements using subsets of available variable data. The methodology follows five general steps: defining relationship structures, setting analysis dimensions and their granularity, selecting the method for fitting the relationship functions, fitting the functions using available data, and quantifying the residual variability. This framework is applied to the Washington D.C. Metro using schedule and operational train movement data to analyse and characterise delay propagation behaviour.
Two model implementations are developed. The first explores the full breadth of variables available in the delay data to identify those offering statistically significant delay relationships, revealing that propagation occurs predominantly between directly connected stations and along shared lines. The second, more targeted model quantifies these relationships, demonstrating that propagation strength is independent of delay magnitude and mostly consistent across different time periods. Further cross-examination revealed minimal sensitivity to operational and network design variables, such as headways and connectivity, suggesting that other variables like scheduling regimes might be the cause of the different observed delay propagation behaviours.
The results highlight that localised delay effects dominate over network-wide influences, suggesting that metro operators can improve service reliability by focusing mitigation efforts on key inter-station connections rather than entire lines. The framework offers a versatile and multiplicative foundation for future applications in delay propagation prediction and analysis.
Deep Reinforcement Learning for Ride-hailing Systems
An experimental study on optimizing matching radius for ride-hailing systems using Deep Reinforcement Learning
Adapting Reinforcement Learning (RL) to optimize the ride-hailing system's matching radius has several difficulties and challenges due to the uncertainties in the real-world rider-hailing market. Traditional approaches are normally static, solving the matching problem at specific times through mathematical models. However, these methods often perform inconsistently when dealing with fluctuating ride-hailing supply-demand relationships, particularly during peak hours. On the other hand, the dynamics and complexity of the ride-hailing market and the ride-hailing environment also make it difficult to model the ride-hailing system. The ride-hailing market is easily affected by many variables, such as weather conditions and local traffic conditions. When quantitatively optimizing the matching radius of the ride-hailing matching system, it is critical to reasonably control irrelevant variables. To address these challenges, this study models the ride-hailing matching problem as a Markov Decision Process (MDP). Based on the defined MDP, a ride-hailing matching simulator is developed. Some assumptions and simplifications are also made to ensure high realism while reasonably controlling irrelevant variables and uncertainties. Multi-replay-buffer Deep Deterministic Policy Gradient (MDDPG) algorithm is then applied to handle the optimization problem of the ride-hailing matching radius. Through the interactions between the MDDPG agent and the developed simulator, feedback rewards are received for the agent to improve the policy. The proposed method is then validated in a case study showcasing the application of the developed simulator and the RL algorithm in a real-world scenario in Austin, Texas. The case study includes an analysis of the current ride-hailing market in Austin, how to apply the simulator based on it, the implementation details of the RL algorithm, and the resulting performance improvements. The results of the case study show that the actions obtained from the proposed method outperform all the baselines in multiple scenarios, highlighting the benefits of using Reinforcement Learning to improve ride-hailing efficiency and user experience.
To conclude, the optimization method proposed in this study applies an advanced Reinforcement Learning approach to the ride-hailing system, successfully improving overall efficiency and user experience. The results of this research demonstrate the potential of Reinforcement Learning in optimizing ride-hailing matching systems, offering a promising direction for further exploration. This study lays a solid foundation for future research to build upon, encouraging the development of more optimization methods with RL technologies that can enhance the effectiveness and adaptability of ride-hailing system in increasingly complex and dynamic environments. ...
Adapting Reinforcement Learning (RL) to optimize the ride-hailing system's matching radius has several difficulties and challenges due to the uncertainties in the real-world rider-hailing market. Traditional approaches are normally static, solving the matching problem at specific times through mathematical models. However, these methods often perform inconsistently when dealing with fluctuating ride-hailing supply-demand relationships, particularly during peak hours. On the other hand, the dynamics and complexity of the ride-hailing market and the ride-hailing environment also make it difficult to model the ride-hailing system. The ride-hailing market is easily affected by many variables, such as weather conditions and local traffic conditions. When quantitatively optimizing the matching radius of the ride-hailing matching system, it is critical to reasonably control irrelevant variables. To address these challenges, this study models the ride-hailing matching problem as a Markov Decision Process (MDP). Based on the defined MDP, a ride-hailing matching simulator is developed. Some assumptions and simplifications are also made to ensure high realism while reasonably controlling irrelevant variables and uncertainties. Multi-replay-buffer Deep Deterministic Policy Gradient (MDDPG) algorithm is then applied to handle the optimization problem of the ride-hailing matching radius. Through the interactions between the MDDPG agent and the developed simulator, feedback rewards are received for the agent to improve the policy. The proposed method is then validated in a case study showcasing the application of the developed simulator and the RL algorithm in a real-world scenario in Austin, Texas. The case study includes an analysis of the current ride-hailing market in Austin, how to apply the simulator based on it, the implementation details of the RL algorithm, and the resulting performance improvements. The results of the case study show that the actions obtained from the proposed method outperform all the baselines in multiple scenarios, highlighting the benefits of using Reinforcement Learning to improve ride-hailing efficiency and user experience.
To conclude, the optimization method proposed in this study applies an advanced Reinforcement Learning approach to the ride-hailing system, successfully improving overall efficiency and user experience. The results of this research demonstrate the potential of Reinforcement Learning in optimizing ride-hailing matching systems, offering a promising direction for further exploration. This study lays a solid foundation for future research to build upon, encouraging the development of more optimization methods with RL technologies that can enhance the effectiveness and adaptability of ride-hailing system in increasingly complex and dynamic environments.
Transfer Towards European Train Travel
A stated choice experiment into the effect of train transfer attributes on long-distance leisure travel choice behaviour
We investigate the impact of TCR on the VTT using datasets from the Netherlands (NL2022) and the United Kingdom (UK2014). We hypothesize that TCR might influence the VTT directly, lead to less consistent choices among respondents, and result in a higher willingness to accept (WTA) among those with TCR compared to those without. Using Multinomial Logit (MNL) and Mixed Logit (ML) models, we analyze the effect of TCR on the VTT, whilst allowing for heteroskedasticity. We also separate TCR interactions into WTA and willingness to pay (WTP) components, while accounting for socio-economic interactions to ensure robust findings across different experimental setups.
Our results show that commuter respondents in the NL2022 dataset with full TCR have a VTT approximately 29% to 45% higher than those without reimbursement. In the UK2014 dataset, the increase is even more pronounced, ranging from 58% to 143%. This effect is significant and consistent in experiments where only time and cost are considered. However, as more attributes are introduced, the difference in VTT becomes insignificant. We found no consistent patterns of scale heterogeneity in the VTT for respondents with and without TCR, leading us to reject the hypothesis that TCR recipients make less consistent choices. Moreover, while the interaction with VTT is more pronounced in the WTA domain than in the WTP domain, the model accounting for this does not outperform a model assuming equal TCR interaction in both domains.
In conclusion, VTT is significantly different for commuters depending on whether they receive travel cost reimbursement. However, these differences cannot be explained by scale heterogeneity or differences in WTA and WTP. We argue that national VTT values should consider travel cost reimbursement, particularly as these values are often based on 2-attribute time-cost experiments. ...
We investigate the impact of TCR on the VTT using datasets from the Netherlands (NL2022) and the United Kingdom (UK2014). We hypothesize that TCR might influence the VTT directly, lead to less consistent choices among respondents, and result in a higher willingness to accept (WTA) among those with TCR compared to those without. Using Multinomial Logit (MNL) and Mixed Logit (ML) models, we analyze the effect of TCR on the VTT, whilst allowing for heteroskedasticity. We also separate TCR interactions into WTA and willingness to pay (WTP) components, while accounting for socio-economic interactions to ensure robust findings across different experimental setups.
Our results show that commuter respondents in the NL2022 dataset with full TCR have a VTT approximately 29% to 45% higher than those without reimbursement. In the UK2014 dataset, the increase is even more pronounced, ranging from 58% to 143%. This effect is significant and consistent in experiments where only time and cost are considered. However, as more attributes are introduced, the difference in VTT becomes insignificant. We found no consistent patterns of scale heterogeneity in the VTT for respondents with and without TCR, leading us to reject the hypothesis that TCR recipients make less consistent choices. Moreover, while the interaction with VTT is more pronounced in the WTA domain than in the WTP domain, the model accounting for this does not outperform a model assuming equal TCR interaction in both domains.
In conclusion, VTT is significantly different for commuters depending on whether they receive travel cost reimbursement. However, these differences cannot be explained by scale heterogeneity or differences in WTA and WTP. We argue that national VTT values should consider travel cost reimbursement, particularly as these values are often based on 2-attribute time-cost experiments.
Investigating the Long-Term Impact of Disruptions on Passenger Travel Behavior Using AFC Data
A Case Study of Washington D.C. Metro Network
Market Structures and Coordination Issues in Passenger Rail Markets
A study on policy decision-making for passenger rail transport in The Netherlands
An efficient simulator is developed in this research to model the spatiotemporal matching processes for ride-hailing and ride-pooling services. Through experiments, the reinforcement learning-based matching strategy is compared against fixed interval and real-time matching strategies. Results indicate that the RL-based strategy outperforms in key metrics such as total waiting time and detour delay. Additionally, Potential-based Reward Shaping effectively addresses the sparse reward problem, further enhancing the model’s learning efficiency. The findings validate that dynamically optimizing matching intervals through reinforcement learning can improve the overall efficiency of ride-hailing services in dynamic supply-demand environments and is scalable and adaptable. ...
An efficient simulator is developed in this research to model the spatiotemporal matching processes for ride-hailing and ride-pooling services. Through experiments, the reinforcement learning-based matching strategy is compared against fixed interval and real-time matching strategies. Results indicate that the RL-based strategy outperforms in key metrics such as total waiting time and detour delay. Additionally, Potential-based Reward Shaping effectively addresses the sparse reward problem, further enhancing the model’s learning efficiency. The findings validate that dynamically optimizing matching intervals through reinforcement learning can improve the overall efficiency of ride-hailing services in dynamic supply-demand environments and is scalable and adaptable.
1. A deep learning model (LSTM-MLP-MDN), composed of Long Short-Term Memory (LSTM), Multilayer Perceptron (MLP), and Mixture Density Network (MDN) architectures, which predicts the probability distribution of traffic demand using historical data.
2. A two-stage stochastic programming model, designed to strategically optimize the locations of charging stations and the initial number of vehicles at each station under demand uncertainty.
The primary goal of this framework is to improve the profitability and operational efficiency of the car-sharing system by optimizing both the location and capacity of charging stations, effectively solving the Charging Station Location Problem (CSLP).
We validate the effectiveness, adaptability, and feasibility of our framework through a comprehensive case study in Manhattan, utilizing historical traffic data to ensure the reliability of the deployment plan. Our results demonstrate that integrating deep learning techniques with stochastic programming significantly enhances both the accuracy of demand forecasting and the consistency of resource allocation in the optimization process for charging station locations. Specifically, the LMSP Framework achieves higher operational efficiency in the short term, as evidenced by superior metrics like Demand Satisfaction Ratio (DSR) and Charging Station Utilization Rate (CSU) compared to traditional methods. This ensures more balanced and efficient resource allocation across different demand scenarios. However, traditional approaches tend to perform better in short-term financial indicators, such as profit and Return on Investment (ROI), as they employ more aggressive resource allocation strategies based on higher demand forecasts. Despite this, the LMSP Framework's focus on operational efficiency positions it as a more viable option for long-term profitability and user satisfaction, offering a sustainable solution for urban mobility.
Furthermore, our analysis provides valuable recommendations for future charging station deployments. These findings have important implications for the planning and operation of electric car-sharing systems, potentially contributing to more sustainable and efficient urban mobility solutions for all stakeholders involved.
...
1. A deep learning model (LSTM-MLP-MDN), composed of Long Short-Term Memory (LSTM), Multilayer Perceptron (MLP), and Mixture Density Network (MDN) architectures, which predicts the probability distribution of traffic demand using historical data.
2. A two-stage stochastic programming model, designed to strategically optimize the locations of charging stations and the initial number of vehicles at each station under demand uncertainty.
The primary goal of this framework is to improve the profitability and operational efficiency of the car-sharing system by optimizing both the location and capacity of charging stations, effectively solving the Charging Station Location Problem (CSLP).
We validate the effectiveness, adaptability, and feasibility of our framework through a comprehensive case study in Manhattan, utilizing historical traffic data to ensure the reliability of the deployment plan. Our results demonstrate that integrating deep learning techniques with stochastic programming significantly enhances both the accuracy of demand forecasting and the consistency of resource allocation in the optimization process for charging station locations. Specifically, the LMSP Framework achieves higher operational efficiency in the short term, as evidenced by superior metrics like Demand Satisfaction Ratio (DSR) and Charging Station Utilization Rate (CSU) compared to traditional methods. This ensures more balanced and efficient resource allocation across different demand scenarios. However, traditional approaches tend to perform better in short-term financial indicators, such as profit and Return on Investment (ROI), as they employ more aggressive resource allocation strategies based on higher demand forecasts. Despite this, the LMSP Framework's focus on operational efficiency positions it as a more viable option for long-term profitability and user satisfaction, offering a sustainable solution for urban mobility.
Furthermore, our analysis provides valuable recommendations for future charging station deployments. These findings have important implications for the planning and operation of electric car-sharing systems, potentially contributing to more sustainable and efficient urban mobility solutions for all stakeholders involved.