Slope reliability analysis using support vector machine

A new methodology for slope reliability analysis using support vector machine (SVM) is proposed. The presented method fits the actual performance function of slope via SVM, by performing deterministic computations at some sampling points designed with uniform design method for training SVM. Then, the reliability index and the design point are obtained using first-order reliability method (FORM) and iterative algorithm. Based on SVM model, the failure probability of slope is calculated using second-order reliability method (SORM) and Monte Carlo simulation (MCS). The accuracy and efficiency of the method are demonstrated by comparing with other methods for two illustrative examples. The results show that sampling and constructing SVM in U-space and evaluating performance function in X-space make the procedure easy to perform reliability analysis involving correlated abnormal distribution variables and ready to do SORM. Comparisons among different methods for two example slopes show that the proposed method is more accurate than FORM and has higher efficiency than MCS. In the proposed algorithm, computations of factor of safety and reliability analysis are separate, which makes the method adaptive for both simple problems having explicit performance function and complicated applications requiring commercial software to calculate the factor of safety.