Bi-model RBDO process based on constructing SVM model by using adaptive support vector clamping method

摘要

Kriging model is an efficient method to solve reliability-based design optimization problem with black box constraints. However, the estimation process of the Kriging model must be based on all sample points, so there still needs a large evaluate time, especially when dealing with high dimensional constraints. To overcome this difficulty, a bi-model RBDO process is proposed in this paper. In the first stage, accurate Kriging model of the constraint is constructed. In the second stage, accurate SVM model is constructed based on the Kriging model by using adaptive support vector clamping method. And the optimal design point of the RBDO problem is calculated based on this SVM model. The results show that the proposed approach is more efficient with necessary accuracy when solving the RBDO problem.