基于改进Kriging模型的主动学习可靠性分析方法

Abstract

Active learning Kriging（ ALK） model is able to only approximate the performance function in a narrow region around the limit state surface.Therefore,the efficiency of reliability analysis is remarkably improved.However,most of the existing strategies build the ALK model based on a so-called DACE toolbox.DACE cannot obtain the global optimal parameter of a Kriging model and the training point chosen in each iteration cannot be the optimal one.In this paper,one famous global optimization,i.e.,the differential evolution algorithm is introduced to explore the optimal parameter of Kriging model and improve the accuracy of Kriging prediction information.As a result,the training point in each iteration is guaranteed to be the global optimal one and the efficiency of ALK model is largely improved.