An Online Approach to Solving Public Transit Stationing and Dispatch Problem

TL;DR

This work addresses dynamic disruptions in fixed-line public transit by formulating stationing and dispatch as a non-myopic online decision problem. It introduces a semi-Markov decision process solved via Monte-Carlo Tree Search, leveraging generative data models for passenger arrivals and disruptions to operate in real time. Across real-world MTA data and synthetic scenarios, the approach improves total passengers served and reduces deadhead miles compared with greedy baselines, while keeping decision times within practical limits. The framework demonstrates the feasibility and value of proactive, non-myopic resource allocation for urban transit systems.

Abstract

Public bus transit systems provide critical transportation services for large sections of modern communities. On-time performance and maintaining the reliable quality of service is therefore very important. Unfortunately, disruptions caused by overcrowding, vehicular failures, and road accidents often lead to service performance degradation. Though transit agencies keep a limited number of vehicles in reserve and dispatch them to relieve the affected routes during disruptions, the procedure is often ad-hoc and has to rely on human experience and intuition to allocate resources (vehicles) to affected trips under uncertainty. In this paper, we describe a principled approach using non-myopic sequential decision procedures to solve the problem and decide (a) if it is advantageous to anticipate problems and proactively station transit buses near areas with high-likelihood of disruptions and (b) decide if and which vehicle to dispatch to a particular problem. Our approach was developed in partnership with the Metropolitan Transportation Authority for a mid-sized city in the USA and models the system as a semi-Markov decision problem (solved as a Monte-Carlo tree search procedure) and shows that it is possible to obtain an answer to these two coupled decision problems in a way that maximizes the overall reward (number of people served). We sample many possible futures from generative models, each is assigned to a tree and processed using root parallelization. We validate our approach using 3 years of data from our partner agency. Our experiments show that the proposed framework serves 2% more passengers while reducing deadhead miles by 40%.

Figures (11)

• Figure 1: (a) An overview of the stationing and dispatch problem. ①Transit agencies have a set schedule for their buses as detailed in GTFS. A subset of buses is held back, to be dispatched in case of incidents, based on some ② decision procedure and some desired reward. Finally, ③ a list of assignments for all buses is generated. (b) Our proposed approach runs inside ② Decision Procedure. It creates generative models based on real-world data and outputs the optimal actions for trip assignments.
• Figure 2: Simulator diagram depicting the main components, passenger arrival model, disruption forecast, and decision agent.
• Figure 3: Parameter search for $\mathcal{C}$ and number of MCTS simulations. Selected parameters are colored red, selected due to the highest average passengers served and run time.
• Figure 4: Comparing the performance of our approach to baselines when both the real world environment and MDP environment are sampled from generative models.
• Figure 5: Comparing deadhead miles of the approach against the different stationing baseline on the real-world data. MCTS_5 is the same approach except decisions are done at 5-minute intervals compared to 15-minute intervals.