The rapid assessment of building damage within a region after an earthquake is crucial for post-earthquake relief efforts. The current building damage assessment methods primarily employ remote sensing or structural equation modeling, which suffer from poor timeliness, are largely focused on individual buildings, and face difficulties in obtaining structural data. Furthermore, building assessment cases are often applicable only to a single earthquake, exhibiting poor generalization performance when the study area changes. This paper addresses the above issues by selecting historical earthquake cases from different geographical regions. The data includes hazard-causing factors, hazard-affected body factors, and hazard-formative environment factors captured by multi-sensors, as well as damage proxy map (DPM) data. In this study, we developed a technical approach to improve the generalization performance of building earthquake damage assessment using the light gradient boosting machine (LightGBM) and sequential least squares quadratic programming (SLSQP) methods. Among them, the LightGBM method is used to construct the evaluation model, while the SLSQP method is used to seek the optimal combination of single-earthquake-case models when constructing a multi-earthquake-case model. The analysis shows that the constructed multi-earthquake-case model is superior to the baseline model. Compared with the baseline model, the constructed multi-earthquake-case model has an mean absolute error (MAE) reduced by 0.015–0.037, an root mean squared error (RMSE) reduced by 0.021–0.056, and an coefficient of determination (R2) increased by 0.096–0.296. Furthermore, the availability of historical earthquake cases as prior data can improve the effectiveness of post-earthquake building damage assessments and is suitable for damage assessments lacking building structural data.
Chen et al. (Sun,) studied this question.