Adaptive and flexible rail transit network service dispatching as a partially observable Markov decision process

Abstract

This paper presents a novel adaptive train scheduling framework with flexible fleet sizes for routing and scheduling in network-wide rail transit services. This framework aims to minimize both passenger waiting times and operating costs driven by prevailing passenger demand. The train scheduling problem is formulated as a partially observable Markov decision process (POMDP) to reflect the practicality in training and real-world applications. To address the computational challenges associated with the train scheduling problem, deep reinforcement learning techniques are applied to seek potential optimal solutions to the optimization problem. The proposed train scheduling framework is tested using real-world scenarios and the data collected from the Hong Kong Light Rail Transit (LRT) network. The experiment results demonstrate that the proposed train scheduling framework using flexible fleet sizes can effectively reduce passenger waiting time and operating costs. This study contributes to the real-time routing and scheduling of network-wide rail transit services by advanced optimization technology.