A novel dynamic-candidate-pool-based sequential metamodel method for slope reliability analysis: Insights from optimization methodology

Sequential metamodel-based slope reliability analysis can significantly reduce the number of calls to a slope stability model, thereby enhancing computational efficiency. However, most existing methods rely on static candidate pools created by random sampling, either locally or globally, which often results in pools that are excessively large and lead to untargeted sampling. To address this issue, this paper proposes a novel dynamic-candidate-pool-based sequential metamodel (DCP-SM) method for slope reliability analysis informed by an optimization methodology. This method optimally utilizes the information provided by a dynamically updated metamodel Ĝ_T. Our recently proposed normal search particle swarm optimization algorithm is utilized to optimize |Ĝ_T|. Solution sets that are evenly distributed and proximal to the predicted limit state function are used to construct a DCP. A novel sequential sampling strategy was proposed to identify the most informative points efficiently by taking full advantage of the characteristics of DCP. The efficacy of DCP-SM was validated by benchmarking it against nine state-of-the-art methods on three explicit performance functions and two typical examples of slope engineering. The results confirm the superiority of DCP-SM in terms of computational efficiency, accuracy, and stability.