Distributed dynamic route guidance via passenger information display systems for subway disruption management

Abstract

Passenger information display systems (PIDS) play a critical role in travel guidance during subway disruptions, but their potential for offering prescriptive route suggestions remains underutilized. Addressing this gap, this study introduces a PIDS-based route guidance framework that employs a distributed guidance approach to manage subway disruptions. This framework leverages the diversion capacity of multiple transfer stations, thereby facilitating network-wide route guidance and mitigating localized congestion. The implementation of the framework involves constructing an evacuation network, where alternative route information is released at evacuation start stations and guides passengers to detour towards evacuation end stations. Only specific transfer stations are selected as these evacuation stations according to an analysis of historical passenger flow distributions. This targeted selection process narrows the optimization space for information release. A dynamic information release optimization problem is formulated, where each pair of evacuation start and end stations is used as a decision variable, with the dual objectives of minimizing travel cost and the number of passengers in the subway. This problem is solved using the asynchronous advantage actor-critic algorithm, which is adept at handling the high-dimensional action and state spaces in a large-scale subway network. This study is the first to integrate PIDS-based route guidance with deep reinforcement learning for optimizing dynamic information dissemination in subway systems. The performance of the proposed framework is validated with data from a subway operation experiencing disruptions. Compared to localized guidance, the proposed framework achieves a 10.87% reduction in total travel cost, a 50.57% greater increase in completed trips, and a 43.83% reduction in peak passenger volume at stations adjacent to the disrupted area.