Seismic risk assessment models for reinforced concrete building clusters considering advanced soft computing algorithms and empirical frameworks

To quickly and accurately estimate the seismic vulnerability and damage states of reinforced concrete (RC) structure clusters, this study utilises multisource artificial intelligence algorithms and classical numerical simulation strategies to develop an intelligent analytical model for predicting structural seismic risk and damage states. Nonlinear time-history and frequency-domain analysis methods are used to analyse monitoring records (932,262 earthquake accelerations) from ten real stations for the Wenchuan earthquake in China and the Kyushu earthquake in Japan. A seismic vulnerability prediction framework for RC structures is proposed that considers nonlinear fitting and a piecewise linear algorithm. A case study is conducted on the proposed prediction framework using a seismic damage dataset (984 RC structures) obtained from real field observations. A seismic vulnerability prediction model based on the random under-sampling boosting (Rusboost) algorithm is developed by combining reinforcement and ensemble learning. The proposed method is compared and analysed with decision tree (DT) and random forest (RF) models. The results indicate that, compared with classical learning algorithms (DT and RF), the Rusboost model achieves relatively high prediction accuracy and efficiency. Using a finite element modelling strategy, a three-dimensional model of a four-story RC structure is established. Structural responsiveness and seismic failure analyses are conducted. A seismic vulnerability comparison model for RC structures that incorporates intelligent and numerical algorithms is developed. The proposed progressive soft-computing and empirical model can serve as a positive reference for the rapid prediction and assessment of the seismic vulnerability and damage states of RC structures.