An efficient surrogate model for reliability analysis of the marine structure piles

Abstract

A hybrid random-forest-based subset simulation (RFSS) method for probabilistic assessment of scour around pile groups under waves is proposed. In the RFSS, a random forest is used to replace the true limit state function (LSF); it is updated based on design samples in the first and last levels of the subset simulation method. For modelling, 127 laboratory datasets collected from the literature were used. First, an existing equation for predicting the scour depth around piles was modified using a metaheuristic approach. The performance of the modified equation was compared with existing equations and models. The modified equation was found to be more accurate than the existing formulas and AI-based models. A probabilistic model based on the RFSS was then developed by considering the modified formula as the LSF of scour depth. Solving two numerical problems, one hydraulic engineering problem and one scour problem validated the robustness and accuracy of the structural reliability method. The results showed that the RFSS is a robust and efficient method for solving high-dimensional real-world problems. Furthermore, compared to the Monte Carlo simulation, the RFSS was able to estimate the reliability index with less computational cost and the same accuracy.