Slope reliability analysis using surrogate models via new support vector machines with swarm intelligence

• This paper presents an intelligent surrogate model based method for system reliability analysis of soil slopes.
• A novel machine learning technique ν-support vector machine is adopted for establishing the surrogate model.
• The hyper-parameters are chosen by particle swarm optimization and artificial bee colony algorithms.
• Monte Carlo simulations are performed via the proposed surrogate model to estimate the system reliability.

Surrogate model methods are attractive ways to improve the efficiency of Monte Carlo simulation (MCS) for structural reliability analysis. An intelligent surrogate model based method for slope system reliability analysis is presented in this study. The novel machine learning technique ν-support vector machine (ν-SVM) is adopted to establish the surrogate model to predict the factor of safety via the samples generated by Latin hypercube sampling. Global optimization algorithms particle swarm optimization and artificial bee colony algorithm are adopted to select the hyper-parameters of ν-SVM model. The applicability of the ν-SVM based surrogate model for slope system reliability analysis is tested on four examples with obvious system effects. It is found that the proposed surrogate model combined with MCS can achieve accurate system failure probability evaluation using fewer deterministic slope stability analyzes than other approaches.