Rapid seismic damage assessment of building portfolio based on fusion of surrogate model and monitored data

Accurate and rapid estimation of regional seismic damage is vital for urban planning, scientific disaster preparedness, and emergency response. However, existing methods face challenges in quickly and accurately obtaining the linear and nonlinear parameters of regional building portfolios with limited building information, making it difficult to realize accurate and rapid regional structural damage assessments. This paper proposes a data fusion method that uses surrogate model and monitored data of building portfolios to rapidly acquire both linear and nonlinear parameters under conditions of limited building information, thereby enabling accurate and rapid assessment of regional seismic damage. To accelerate structural response calculations, the best surrogate model was obtained by comparing three Machine Learning (ML) algorithms tuned by the Bayesian Optimization (BO) algorithm. Predicted responses by the surrogate model, and monitored responses were then fused using Particle Swarm Optimization (PSO) and parallel computing to inverse the structural parameters. Finally, the target ground motions and the obtained structural parameters were input into the surrogate model for regional seismic damage assessment. To validate the proposed method, two cases with different sources of monitored data were presented. The results show that the errors between the predicted and actual values in structural damage state proportions under new ground motions for both two cases are within 5 % and the structural parameters closely match the actual situations. Additionally, the regional data fusions for various scales of building portfolios were conducted. The calculation speed improved by at least 5880 times faster compared to existing methods for different scales of building portfolios, demonstrating the capability of the proposed method to rapidly assess large-scale building portfolios.