Slope reliability analysis considering spatial variability of parameters based on efficient surrogate model

When evaluating the slope reliability with small failure probability considering the spatial variability of parameters, the traditional reliability analysis methods are often time-consuming or difficult to solve. Thus, it is aimed to propose an active learning multiple adaptive regression spline method based on the segmental inverse regression combined with the subset simulation for the slope reliability analysis, and the effects of the sliced inverse regression method on the response values of the surrogate model are explored. Firstly, the random field of shear strength parameters is discretized by the Karhunen-Loève expansion method, and a certain number of training samples are calculated by the strength reduction method to establish the initial surrogate model. Then, the active learning function is used to select a certain amount of optimal sample points to update the surrogate model, and the final surrogate model is coupled with the subset simulation for the slope reliability analysis. Finally, the effectiveness of the proposed method is verified by two typical spatial variation slope examples. It is shown that the proposed method can obtain more accurate results and avoid memory overflow by using fewer training samples for the slope reliability analysis with small failure probability.