Gaussian process-based dynamic response surface method for estimating slope failure probability

In light of the limitation of the traditional response surface method for slope reliability analysis with high nonlinear implicit performance function, Gaussian process regression (GPR) model, which is a capable of solving the highly nonlinear regression problem with small samples and high dimensions, is applied to rebuild response surface of implicit performance function. Basing on the optimum sample selected by the function of uncertainty evaluation of GPR model, an iterative algorithm is presented to update GPR response surface self-adaptively. Thus, a new method combing GPR based dynamic response surface with Monte Carlo Simulation (MCS) method for slope reliability is proposed. The accuracy and feasibility of the presented method are demonstrated by numerical examples. The reliability of three slopes are analyzed using the presented method. The results show that the proposed method has higher efficiency and higher accuracy compared to traditional response surface method. It could be achieved easily and can directly take advantages of existing slope analysis codes without any modification. Thus, the proposed method provides a powerful tool for fast analysis of slope reliability.