Real-Time Seismic Damage Prediction and Comparison of Various Ground Motion Intensity Measures Based on Machine Learning

After earthquakes, an accurate and efficient seismic-damage prediction is indispensable for emergency response. Existing methods face the dilemma between accuracy and efficiency. A real-time and accurate seismic-damage prediction method based on machine-learning algorithms and multiple intensity measures (IMs) is proposed here. 48 IMs are used for representing the ground-motion characteristics comprehensively, and the workload of the nonlinear time-history analysis (NLTHA) method is replaced by model training in the non-urgent stage to promote efficiency. Case studies with various buildings prove the accuracy and efficiency of the proposed method, and corresponding key IMs are identified by iterative optimization.