TL;DR
This paper introduces SimFLEX, a novel simulation framework for evaluating the feasibility and performance of on-demand feeder bus services in urban areas. By integrating microsimulation, traveler learning models, and public transport routing, SimFLEX allows for data-driven comparisons of potential service locations, even with unknown demand, offering a crucial tool for urban planners.
What Problem Does This Paper Address?
Urban areas are increasingly looking to on-demand feeder bus services to improve public transport accessibility, especially in remote or underserved neighborhoods. However, planning these services is challenging due to the difficulty in predicting demand and user behavior before launch. Existing methods often lack a comprehensive framework for evaluating service feasibility across different urban contexts, leading to potential inefficiencies and failures. This paper addresses this gap by providing a methodology that simulates and compares the performance of on-demand feeder services in various urban districts, enabling informed decisions about where to deploy them for maximum impact.
Methods
SimFLEX employs a multi-faceted approach:
- Microsimulation and Demand Generation: It disaggregates macro-level origin-destination data into synthetic agents to create diverse demand scenarios, capturing individual traveler behavior variability.
- Discrete Choice Modeling: A binary logit model is used to simulate traveler mode choice between the integrated feeder service and conventional public transport, considering factors like travel time, cost, and convenience.
- Traveler Learning (MSA): The Method of Successive Averages (MSA) is used to simulate how travelers learn and adapt their behavior over time, leading to system stabilization and more realistic travel time estimates. This is crucial for understanding long-term performance.
- Ride-Pooling Optimization (ExMAS): The ExMAS framework is integrated to compute shared trip attributes and operational KPIs for the feeder buses, optimizing ride-pooling efficiency.
- Public Transport Routing (OpenTripPlanner - OTP): OTP is used to integrate feeder services with existing public transport networks, enabling the evaluation of the entire multimodal system.
- Key Performance Indicators (KPIs): SimFLEX calculates a suite of KPIs, including feeder attractiveness, reduction in waiting time, overall added value, vehicle occupancy, and changes in vehicle and passenger hours.
- Sensitivity Analysis: The framework allows for sensitivity analysis of KPIs to variations in alternative-specific constants (ASCs), which represent unobserved factors influencing mode choice, thus assessing the robustness of the results.
Datasets
The study utilizes data from Krakow, Poland, including:
- Spatially defined urban areas (polygons).
- Public transport hubs locations.
- City population distribution data.
- City zones (multipolygons).
- Origin-Destination (OD) matrices (based on 2017 data, adjusted for 2023).
- General Transit Feed Specification (GTFS) data for transit schedules.
- City road network graph (from OpenStreetMap via OSMnx).
Key Insights and Ideas
- The importance of traveler learning: SimFLEX highlights that traveler behavior adapts over time, with a simulated two-week period (30 MSA iterations) showing stabilization of expected travel times. This emphasizes that initial simulations might not reflect long-term adoption.
- Skotniki as a more promising candidate: Across most key metrics, Skotniki consistently shows better potential for on-demand feeder service implementation compared to Bronowice.
- Trade-offs in service benefits: While Skotniki offers higher attractiveness and added value, Bronowice demonstrates a significantly greater potential for reducing public transport waiting times. This highlights that the "best" location depends on the specific goals of the transportation authority.
- Robustness of findings: The sensitivity analysis across a range of ASC values confirms Skotniki's superiority, indicating that the results are not overly dependent on specific assumptions about unobserved factors.
- Addressing the "unknown demand" problem: SimFLEX's ability to estimate demand and user preferences without prior real-world data is a significant contribution, enabling pre-deployment evaluation.
Key Results
- Comparative Analysis: Skotniki showed a higher probability of feeder service choice (0.265 vs. 0.248), significantly higher feeder attractiveness (-1.966 vs. -2.690), and higher overall added value (0.600 vs. 0.568) compared to Bronowice.
- Waiting Time Reduction: Bronowice offered a substantially greater reduction in waiting times (65.904 seconds vs. 28.925 seconds), a nearly 77% improvement over Skotniki.
- Operational KPIs: Bronowice exhibited slightly higher average vehicle occupancy and increased vehicle-hours traveled, while Skotniki showed a larger increase in passenger-hours traveled, indicating a potential trade-off between vehicle utilization and passenger travel time.
- Sensitivity Analysis: The superiority of Skotniki was consistently observed across all tested ASC variations, reinforcing the reliability of the findings.
Limitations and Future Work
- Limited Transportation Options: The current analysis considers only two options: integrated feeder or public transport. Future work could incorporate private car usage, ride-sharing, and micromobility.
- Traveler Heterogeneity: The study could be enhanced by investigating how traveler heterogeneity (e.g., value of time, ASC) impacts simulation outcomes.
- Optimal Area Determination: SimFLEX's computational power could be leveraged to determine the optimal service area rather than just comparing pre-selected ones.
- Socio-demographics and Trip Characteristics: Incorporating detailed socio-demographic data and trip characteristics (e.g., age, trip purpose) could refine predictions.
- Broader Urban Contexts: Expanding the evaluation framework to a wider range of urban areas with diverse characteristics would improve generalizability.
- Advanced Learning Models: Exploring more sophisticated learning models beyond MSA could further enhance the simulation's realism.
Summary updated