Modelling seismic fragility of a rock mountain tunnel based on support vector machine

Abstract

In the paper, an analytical method is proposed to develop seismic fragility analysis for rock mountain tunnels. We consider four types of uncertainties in the fragility analysis including different ground motions, tunnel depths, rock mass and lining thickness. By using the uniform design method (UDM), numerical experiment samples are generated. The verified dynamic numerical simulation (DNS) model is carried out to develop probabilistic seismic demand models. To optimize conventional methodology, a prediction technique support vector machine (SVM) is employed. The SVM model could help to reduce calculation resource. It is concluded that (1) the proposed uniform design-dynamic numerical simulation-support vector machine (UDM-DNS-SVM) method could provide accurate estimated fragility curves considering multiple uncertainties; (2) comparisons among the proposed fragility curves, case studies and empirical curves verified feasibility of proposed fragility curves.