Cross-system modeling and analysis of cascading failure propagation in large-scale metro stations under extreme flooding events

Climate-driven intensification of extreme precipitation is significantly increasing unprecedented flooding risks for underground transportation infrastructure. Current flood risk assessment approaches for large-scale metro stations, characterized by extensive subsystem integration, high passenger volumes, and complex spatial configurations, often fail to capture critical cross-system interactions. This limitation arises mainly from the artificial separation of simulations using computational fluid dynamics (CFD) from cross-system network analysis, which hinders the accurate prediction of cascading infrastructure failures. This study develops an integrated framework that combines CFD simulations with multi-layer network theory to concurrently analyze flood dynamics and system interdependencies. This framework models four critical subsystems of large-scale metro stations, including power, drainage, communication, and pedestrian, as interconnected networks based on established engineering standards. Flood-depth-dependent functions determine infrastructure node states, with thresholds calibrated from engineering standards to ensure physical consistency. The validations with the Shanghai Eastern Hub during a 500-year rainfall event (327 mm over 6 h, with a peak intensity of 90 mm/h) demonstrate that power systems exhibit the highest vulnerability, with functionality declining to 51.5% within 60 min. Furthermore, an analysis of 13,241 cascading failure events reveals that 67.3% are driven by water depth, while 32.7% are influenced by inter-system dependencies. Network analysis uncovers a critical importance-vulnerability paradox: power systems, serving as the network backbone with the highest importance score (0.446), simultaneously exhibit disproportionately elevated vulnerability (0.172) compared to other subsystems. The developed framework incorporates minute-level temporal coupling, validated to capture the dominant characteristics of infrastructure responses while maintaining computational tractability for engineering applications.