基于物理信息增强的高层结构地震反应智能重构与预测OA

High-rise buildings,as critical carriers of urban functionality,inevitably experience structural performance degradation during long-term service,posing challenges for accurate seismic performance assessment.Structural health monitoring(SHM)provides an essential means to capture in-service structural conditions.However,its effectiveness is often constrained by sparse measurements and missing data,which hinder comprehensive full-field response evaluation.Consequently,there is an urgent need for methods capable of reconstructing and predicting structural seismic spatio-temporal fields under such incomplete observation conditions.To address this issue,this study proposed a physics-enhanced framework for multi-step spatiotemporal prediction of seismic responses in high-rise buildings.The framework represented floors as graph nodes,constructed adjacency matrices based on identified mode shapes and inter-story dynamic coupling,and employed a spatiotemporal interaction graph neural network to enable response reconstruction and multi-step prediction under sparse observations.Meanwhile,the hyperparameters sensitivity analysis was performed to reveal the correlations between the hyperparameters with the prediction accuracy.The proposed method was systematically validated using the Millikan Library at the California Institute of Technology,USA,and a landmark supertall building in Shanghai,China.Results demonstrate that the framework accurately reconstructs responses at uninstrumented floors and effectively captures the propagation of vibration information across nodes in both time and frequency domains.Across multiple scenarios,the normalized root-mean-square error remains below 5%,highlighting the high prediction accuracy and strong engineering applicability of the proposed approach.