A clustering-based analysis method for simulating seismic damage of buildings in large cities

Abstract

Earthquakes are major threats to cities. Refined simulations of urban earthquake damage are significant for disaster prevention planning, including post-earthquake emergency rescue operations. However, this approach is time-consuming and computationally demanding. This study presents a novel machine learning method for simulating seismic damage of large urban building groups, thus minimizing the computational time and resources associated with the simulation process. First, the proposed method clusters building structures based on their seismic damage index (DI) under specific ground motions. Subsequently, one structure is selected from each cluster for seismic analysis. This approach significantly reduces the number of analyzed structures. The efficiency and accuracy of this method are studied based on a parametric study that involved an improved K-means clustering algorithm and a grid-based clustering algorithm. A batch-clustering algorithm is also developed to further improve the speed of clustering and efficiency of regional seismic simulations. A large Chinese city is considered as research application example. The results indicated the following: (1) the method based on the improved K-means clustering algorithm was superior; (2) the batch-clustering algorithm significantly speeded up clustering analysis and enhanced efficiency of the seismic simulation; (3) the clustering-based simulation method demonstrated high efficiency and accuracy, with a significant decrease of 90.5% in the calculation time compared with direct nonlinear time history analysis. In addition, the average relative error of the DI was only 11.0%. Moreover, over 85.0% of the structures were estimated to be in the correct damage state.