Optimum Scheme Selection for Multilayer Perceptron-Based Monte Carlo Simulation of Slope System Reliability

Abstract

Surrogate models are helpful tools to enhance the efficiency for intensive computations of the factor of safety (FoS) in probabilistic slope stability evaluation. This study presents a multilayer perceptron (MLP)-based surrogate model combined with the Monte Carlo simulation (MCS) for system reliability analysis of earth slopes. The MLP-based surrogate model is constructed derived from the space-filling Latin hypercube sampling (LHS) for a global prediction of the FoS. Several factors affecting the performance of the MLP model are studied in detail, including the training algorithm, the generation method and size of samples, and the hyperparameters. Three examples with system effects are tested to verify the performance of the proposed method. The results show that the MLP-based MCS can achieve high accuracy and efficiency for the system failure probability assessment of soil slopes in different failure zones.