Enhancing fixed transit services with demand-responsive limited-stop services considering alternative routes

• Main Authors: Lee, Kelvin, Jiang, Yu, Dauwels, Justin, Su, Rong
• Other Authors: Interdisciplinary Graduate School (IGS)
• Format: Conference or Workshop Item
• Language: English
• Published: 2024
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence; Engineering::Computer science and engineering::Computing methodologies::Simulation and modeling; Engineering::Civil engineering::Transportation; Public Transport; Limited-Stop Service; Transit Assignment; Data-Driven Optimisation; Reinforcement Learning.
• Online Access: https://hdl.handle.net/10356/171866; https://www.hksts.org/conf23.htm
• Institution: Nanyang Technological University

The rigidity of conventional bus services that do not adapt to real-time changes in demand or travel times can lead to inefficient use of limited resources because these services are planned based on historical information, which may not be representative of the actual situation. This study proposes a model that makes the fixed services more demand responsive by reallocating some buses from the existing service to a new limited-stop service that operates alongside the existing service in real-time based on observed or predicted travel demands. Contrary to the existing literature on stop-skipping for bus services, the proposed limited-stop service is not subject to any service pattern limitations that determine whether a stop can be skipped. Furthermore, this study allows buses to detour or reroute to another shorter route instead of traversing the original routes and passing the skipped stops without stopping. By considering shorter alternatives, the proposed service uses available resources more efficiently and could reduce passengers’ travel times. To solve the model for real-time applications, a reinforcement learning-based solver is developed to efficiently obtain high-quality solutions for arbitrary instances of the problem with varying sizes, travel times, and travel demands. The proposed approach is evaluated on 30 real-world transit routes and is found to reduce total passenger travel time by an average of 2.6%, with potential reductions as high as 16.3%. The proposed solver method is comparable to tabu search on average across ten runs, but it can outperform tabu search by up to 9.7% in individual runs.