Visual support vector machine-based method for reliability-based design optimization

摘要

In this paper, a method based on Virtual Support Vector Machine (VSVM) and Kriging response surface model is proposed to address Reliability-Based Design Optimization (RBDO) problems for black-box system. In many engineering optimization problems, the number of function evaluations is limited because of the expensive evaluation of the objection function and constraint functions. To get a more precise result, it is necessary to evaluate more samples in a small region, which includes the optimum point. Considering these characters, the RBDO method based on VSVM (RBDO-VSVM) is proposed in this article. RBDO-VSVM uses VSVM classification to approximate the limit state function, which can guide the constraint boundary sampling process. Besides, Trust Region (TR) method is used in the algorithm to downsize the search space to a local range around the constraint boundary, which probably contains the optimum point. Numerical optimization example is posed in the end of the article to validate the efficiency and accuracy of the proposed method.