Joint rescheduling for timetable and platform assignment of High-Speed Railways via graph neural network-based Deep Reinforcement Learning

Xuan Liu, Min Zhou and Hairong Dong

Transportation Research Part E: Logistics and Transportation Review, 2025, vol. 202, issue C

Abstract: Virtual coupling enables trains to travel in convoys with reduced headway, significantly enhancing operational efficiency. Joint rescheduling for High-Speed Railways (HSRs) timetable and platform assignment allows for more efficient utilization of station resources, thereby maximizing the benefits of virtual coupling. This study analyzes the operation processes of high-speed trains occupying routes and platforms under virtual coupling, and develops a joint rescheduling model for timetable and platform assignment under virtual coupling, considering a detailed decomposition of station structures. A Deep Reinforcement Learning (DRL)-based method is proposed to solve the joint rescheduling problem. The state of track occupancy by trains is represented using a heterogeneous graph. Based on this graph, a Markov Decision Process (MDP) is designed according to the constructed joint rescheduling model, achieving platforming and timetabling through track assignment, enabling end-to-end rescheduling. Graph Neural Networks (GNN) integrated with an attention mechanism are employed to effectively address the challenge of applying trained policies to train rescheduling problems of varying scales. The GNN efficiently captures node and edge features within the heterogeneous graph, resulting in size-agnostic performance. The numerical experiments are conducted based on real data from the Beijing–Shanghai High-Speed Railway. The proposed method can reduce total train delay by an average of 9.8% compared to the commonly used scheduling rules. It also shows high solving efficiency and stability compared to CPLEX and Genetic Algorithm (GA). Moreover, the solution time grows approximately linearly with the problem size. In particular, the learned policies can still achieve good results when solving large-scale and cross-line rescheduling problems that have not previously been encountered, demonstrating strong generalization capabilities.

Keywords: Timetable; Platform assignment; Virtual coupling; Heterogeneous graph; Graph neural network (search for similar items in EconPapers)
Date: 2025
References: Add references at CitEc
Citations:

Downloads: (external link)
http://www.sciencedirect.com/science/article/pii/S1366554525003187
Full text for ScienceDirect subscribers only

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:eee:transe:v:202:y:2025:i:c:s1366554525003187

Ordering information: This journal article can be ordered from
http://www.elsevier.com/wps/find/journaldescription.cws_home/600244/bibliographic
http://www.elsevier. ... 600244/bibliographic

DOI: 10.1016/j.tre.2025.104277

Access Statistics for this article

Transportation Research Part E: Logistics and Transportation Review is currently edited by W. Talley

More articles in Transportation Research Part E: Logistics and Transportation Review from Elsevier
Bibliographic data for series maintained by Catherine Liu ().