Enhanced Active Learning for Structural Reliability Analysis: An Ensemble SVR Metamodel‐Monte Carlo Approach

ABSTRACT

The article introduces a novel active learning method for structural reliability analysis EMSVR‐MCS, which combines the ensemble of metamodels (EMs) method, support vector regression (SVR) metamodel, and Monte Carlo simulation (MCS), this approach offers an effective and efficient solution for evaluating failure probability. The method proposes constructing a robust average metamodel from a set of SVR metamodels, each trained with a different kernel function. Additionally, it introduces the construction of a variance to estimate the prediction error of the average metamodel by exploiting differences among the predicted responses of each metamodel. The method also uses a new approach to select points for training, ensuring they cover the design space effectively using an innovative learning function. Five examples of varied nature and dimension were used to validate this method, demonstrating its effectiveness in estimating failure probability while significantly reducing computational costs compared to other approaches.