Support vector machine response surface method for structural reliability analysis

In order to predict failure probability of implicit performance function for complex structures, a support vector machine response surface method (SVMRSM) is presented to alleviate the computational effort. The iterative concept of the proposed method is similar as that of classical response surface method (RSM). The support vector regression (SVR) is employed to establish the surrogate of the implicit performance function in SVMRSM; however, the rigid polynomial is employed in the classical RSM. SVR model is then connected to the first order reliability method (FORM) to calculate reliability index and form an iterative process. Moreover, an improved method is proposed for the selection of training samples, by which the efficiency of SVMRSM is improved further. Several comparisons using explicit or implicit performance functions are presented to illustrate the proposed method for structural reliability analysis. The SVMRSM is found to yield higher accurate results and to be most efficient.