Adaptive ensemble learning of radial basis functions for efficient geotechnical reliability analysis

Abstract

Surrogate models can provide an effective tool to improve the computational efficiency for geotechnical reliability analysis when direct evaluations of the original performance function (OPF) of a geotechnical structure are time-consuming. However, selection of a proper surrogate model is often problem dependent. A particular surrogate model may not suitable for different geotechnical structures. Meanwhile, training of the surrogate model sometimes can also cost significant computational time. To address these issues, this paper proposes an efficient and robust geotechnical reliability analysis method based on the adaptive ensemble model of radial basis functions (AERBF) and Monte Carlo simulation (MCS). The ensemble of multiple radial basis functions (RBFs) is constructed to replace the OPF of a geotechnical system, which can improve the robustness of the surrogate model. The ensemble model does not need to select a certain RBF but automatically determines the weight coefficients of each component RBF through the training process, which can avoid the possible selection of an improper RBF. An active learning function is adopted to select the most suitable training samples to train the ensemble model, which can improve the efficiency significantly. Several representative examples are illustrated to validate the accuracy and efficiency of the proposed method. The results show that the approach is robust to different kinds of geotechnical systems. It can accurately estimate the failure probability of geotechnical systems with a small number of evaluations of the OPF.