Abstract:
To address the issues of high computational cost of Monte Carlo simulation, insufficient convergence efficiency of the AK-MCS method when dealing with complex problems, and the limited capability of traditional learning functions, this paper proposes a structural reliability analysis method based on adaptive ensemble Kriging. The proposed method first employs a high-density initial sampling strategy based on space-filling mechanisms, incorporating local void detection and region-filling schemes to enhance the overall coverage of the design space and improve the accuracy of the initial Kriging surrogate model. Furthermore, a novel L-learning function is constructed by integrating the probability of prediction error, sample spatial distance, and a dynamic exploration factor. Meanwhile, an ensemble surrogate model is established, in which fusion weights are dynamically computed based on leave-one-out cross-validation errors, and the prediction results of multiple sub-models are integrated to enhance the accuracy and stability of reliability analysis. The effectiveness of the proposed method is validated through reliability analysis of two case studies, a four-failure-domain series system and a six-dimensional nonlinear oscillatory system, and compared with five other reliability analysis algorithms. The results demonstrate the efficacy of the proposed method, providing a feasible solution for structural reliability assessment of complex systems.