Abstract: In highly urbanized soft-soil regions, underground spaces face substantially elevated disaster risks due to the combined effects of weak geological strata, confined environments, and high-density functional layouts. Once initiated, such disasters tend to propagate through chain reactions and cascading failures, forming complex, coupled disaster-chain networks. To improve disaster identification and prevention capabilities, this study integrates complex network theory with a graph convolutional network(GCN)approach to construct an evolution model of disaster-chain networks in urban underground spaces(UUS)in soft-soil areas. The model systematically identifies critical nodes and key propagation paths and evaluates network robustness. Based on 206 UUS disaster cases, a multi-hazard coupled disaster-chain network comprising 115 nodes and 190 directed edges was established. Structural features and centrality metrics of nodes and edges were extracted to characterize the network topology. A three-layer GCN model was developed to intelligently identify critical nodes, outperforming traditional centrality-based methods in accuracy. An edge vulnerability index was also proposed to identify high-risk propagation paths and assess network robustness under simulated failure scenarios. The results indicate that node-based attacks cause significantly greater damage to network connectivity and efficiency than edge-based attacks. Nodes associated with the coupled action of soft soil and groundwater—such as water inrush, pipe bursts, ground collapse, and foundation pit collapse—typically occupy central transmission paths or act as network hubs. Their failure leads to multi-path disconnection and network fragmentation, highlighting their role as dominant control points in the disaster-chain network. Building on the identification of key nodes and paths, this study proposes an integrated framework of "critical node identification-path intervention-propagation control," offering a theoretical basis and technical pathway for risk assessment, propagation analysis, and targeted disaster mitigation in soft-soil urban underground spaces.