A real-time multi-node structural response prediction and rapid seismic resilience assessment method

Earthquake disasters can result in substantial damage to building complexes, highlighting the necessity for efficient methodologies in evaluating structural losses and assessing seismic resilience. In order to improve the efficiency of such assessments, this paper introduce a real-time seismic response prediction method for structures using a deep learning framework, featuring the newly developed MN-Seisformer model. This model innovatively predicts the response of multiple structural nodes with high precision by extracting temporal and nodal features. Coupled with this prediction model, we propose a rapid structural loss evaluation and resilience assessment method that relies on quantile response predictions to provide accurate and reliable evaluations. Our method's effectiveness is demonstrated on a 21-story reinforced concrete building, showcasing the ability to achieve simultaneous multi-node predictions with remarkable accuracy and a prediction speed surpassing traditional finite element calculation by orders of magnitude. The effectiveness and robustness of the MN-Seisformer model are reinforced through ablation studies, while comparative experiments establish its state-of-the-art performance, representing a notable advancement in rapid and precise loss evaluation and resilience assessment. The contributions of this study greatly enhance post-earthquake decision-making and emergency response efficiency.