Integrated train timetabling and rolling stock rescheduling for a disturbed metro system: A hybrid deep reinforcement learning and adaptive large neighborhood search approach

Abstract

For a metro line with high density and short trip time, the original train timetable and rolling stock circulation face infeasible risks under unexpected disturbances. This paper focuses on integrated train timetabling and rolling stock rescheduling for a disturbed metro system to reduce the negative impact. The problem is formulated as a multi-objective master model, in which required operational constraints and dispatching strategies are taken into account. To satisfy the real-time requirement, an innovative solution framework consisting of an independent rescheduling process and a cooperative rescheduling process is proposed. In the independent rescheduling process, the master model is decomposed into a series of submodels in the unit of train service. The submodel can be transformed into a Markov Decision Process (MDP) with well-defined fundamental elements (i.e., state, action, and reward function). Based on the MDP, a hierarchical policy is developed by introducing deep reinforcement learning to generate the initial solution, including the lower-level policy for train timetable and the higher-level policy for flexible rolling stock circulation. In the cooperative rescheduling process, the selfishness of agents that appears after the model decomposition is overcome by an adaptive large neighborhood search algorithm, which can improve the solution quality. Finally, two sets of numerical experiments are conducted to demonstrate the performance of the proposed solution framework. The experimental results show that a near-optimal solution can be obtained in a short time, which is better than the currently used practical rules in the automatic train supervision system, especially during peak hours. Furthermore, the effects of different parameter settings are analyzed.