Digital twin-based resilience analysis and emergency maintenance with generative AI of smart urban metro systems

With the rapid expansion and growing complexity of urban metro systems, operational disturbances increasingly threaten system stability and service continuity. However, existing recovery strategies remain insufficient in terms of emergency response efficiency and intelligent decision-making. To address these challenges, this study proposes a digital twin metro system architecture with generative AI enabling real-time system status monitoring, and gives a new resilience model to optimize the emergency maintenance decision-making.. Furthermore, the study investigates performance and cost variations across failure and recovery phases, and introduces a cost importance-based recovery strategy that enhances system resilience by optimizing the station recovery sequence. Finally, using the Beijing metro as a case study, the impact of different recovery strategies is evaluated under both single-line section and multi-line section attack scenarios. The results indicate that the cost importance-based strategy (CIBRS) outperforms traditional approaches (BBRS, DBRS, ECBRS, RGBRS). In the single-line section attack, it achieves a resilience value of 0.747, with improvements of 74.94%, 35.34%, 95.55%, and 1.91%, respectively. Under the multi-line section attack, the resilience value reaches 0.777, with corresponding improvements of 79.45%, 46.33%, 85.44%, and 5.00%. The result confirms the superior adaptability and robustness of the proposed method in complex scenarios. This study offers valuable insights for intelligent metro system management and emergency maintenance.