Abstract:

With the increasing interplay among multiple hazards and the growing interconnectedness of systems， the operational risks faced by urban critical infrastructures （UCIs） have become increasingly prominent. Existing studies often overlook the nonlinear interactions among hazards and offer limited modeling capabilities for failure processes in multi-interdependent systems under multi-hazard coupling scenarios. To address this gap， a two-stage research framework is propased for failure risk assessment of UCIs under multi-hazard coupling scenarios. First， based on historical disaster data and spatial information， a multi-hazard probabilistic model is constructed， and representative multi-hazard coupling scenarios are generated using a temporal hypergraph method. Subsequently， a high-order topological dynamic graph neural network （HoT-DGNN） model is developed to capture node interactions and high-order network topological features within interdependent systems， integrating multi-source information to effectively predict system failure modes. Finally， an empirical analysis is conducted on the power-gas interdependent system in a typical coastal city. Results demonstrate that the proposed approach can accurately assess the failure risk probabilities of UCIs under multi-hazard coupling scenarios and effectively identify vulnerable components. It enriches the methodology for modeling complex interactions among multiple hazards and interdependent systems， providing both theoretical foundations and practical tools for quantifying systemic risks under multi-hazard scenarios. It offers valuable insights for improving infrastructure resilience and advancing resilient city construction.

Key words: urban critical infrastructures, failure risks, multi-hazard coupling, graph neural network, dynamic modeling