Dynamic recovery sequence optimization of subway-bus network with constrained reinforcement learning

Subway and bus networks constitute the backbone of urban public transportation and cater to the daily commuting needs of city residents. For damaged subway-bus networks caused by extreme events, aligning the repair sequence with dynamic passenger flow is crucial to reducing passenger delays. Most existing studies on station recovery strategies ignore the dynamic changes of passenger flow, and this study aims to design the station recovery sequence for the subway-bus network considering dynamic passenger flow. To enhance practicality, a constrained reinforcement learning framework is formulated to incorporate passenger demands, mobility distance, and repair duration into a Markov decision process. Moreover, dynamic graph convolutional networks are adopted to capture time-varying characteristics of passenger flow. The reward function proposed in this work integrates node importance, recovery progress, transferred passengers, and excessive distance penalty to enable efficient post-disaster recovery. Empirical analysis reveals that this methodology increases the number of transferred passengers by 5.53%-11.30%.

• Record URL:
  • https://doi.org/10.1016/j.trd.2026.105238
• Record URL:
  • http://www.sciencedirect.com/science/article/pii/S1361920926000313
• Availability:
  • Find a library where document is available. Order URL: http://worldcat.org/issn/13619209
• Supplemental Notes:
  • © 2026 Elsevier Ltd. All rights are reserved, including those for text and data mining, AI training, and similar technologies. Abstract reprinted with permission of Elsevier.
• Authors:
  • Liu, Shan
  • 0000-0002-4385-2487
  • Meng, Zhenhao
  • 0009-0003-1199-1495
  • Jiang, Rui
  • 0000-0002-3866-5388
  • Wang, Zhengli
  • 0000-0003-4268-9195
  • Zhang, Ya
  • 0000-0002-7555-162X
  • Ge, Ying-En
  • 0000-0002-8435-0483
• Publication Date: 2026-5

Language

• English

Media Info

• Media Type: Web
• Features: References;
• Pagination: 105238
• Serial:
  • Transportation Research Part D: Transport and Environment
  • Volume: 154
  • Issue Number: 0
  • Publisher: Elsevier
  • ISSN: 1361-9209
  • Serial URL: http://www.sciencedirect.com/science/journal/13619209

Subject/Index Terms

• TRT Terms: Bus transit; Machine learning; Networks; Neural networks; Optimization; Passenger traffic; Public transit; Subways; Weather
• Subject Areas: Passenger Transportation; Public Transportation; Security and Emergencies;

Filing Info

• Accession Number: 01979590
• Record Type: Publication
• Files: TRIS
• Created Date: Feb 17 2026 1:11PM