Spatial Propagation Modeling and Assessment of Seismic Vulnerability of Urban Bridges Based on Graph Neural Network

In view of the problem that urban bridge groups are prone to regional failure due to cascading effects in earthquakes, and traditional vulnerability assessment is difficult to quantify the spatial propagation mechanism, this study proposes a modeling method based on graph neural network (Graph Neural Network, GNN). First, the urban bridge network is abstracted into a graph structure; then, a multi-level GNN model is constructed to integrate topological connection characteristics and ground motion intensity parameters; finally, the spatial diffusion process of vulnerability under earthquake loads is simulated by a dynamic propagation algorithm. The experimental results show that in the bridge failure prediction experiment based on historical earthquake data, the model can better reflect the seismic vulnerability of different bridge types. The regional system resilience analysis experiment further verifies the advantages of GNN in multi-node systems. In the comparative experiment with other models, GNN is superior to other methods in terms of accuracy (87%) and precision (0.85). In summary, the GNN model provides a new and more precise spatial propagation modeling method for seismic vulnerability assessment of urban bridge groups, which has strong practical application potential.