Seismic vulnerability assessment model of civil structure using machine learning algorithms: a case study of the 2014 Ms6.5 Ludian earthquake

Constructing a seismic vulnerability assessment model is crucial for emergency response and mitigating seismic risk. In the wake of increased usage of machine learning, it helps the assessment model to produce more spatial and quantitative results with higher speed, providing a global view of seismic vulnerability in large areas. This paper explores the applicability and performance of different machine learning algorithms in vulnerability assessment. Three regression algorithms of supervised learning (multiple linear regression, BP neural network and support vector machine) are adopted for testing and validation of the models. The detailed vulnerability inventories of different categories of civil structures under Ludian earthquake are compiled according to field investigation and adopted as the training set. Various kinds of damage ratio are determined to consider the different degrees of impact on civil structures under earthquake. Four aspects of factors containing structure category, topographical, seismic, comprehensive factors are selected as independent variables. It is found that the prediction error shows distinctions in different categories of civil structures irrespective of the algorithms, suggesting the necessity of considering structure category in seismic vulnerability assessment using machine learning. Nonetheless, the highest success rates are obtained using BP neural network, with an RMSE of less than 0.05.