关键词: 响应面法, 模型修正, 最小二乘支持向量机, 核函数, 损伤识别

Abstract:

According to the excellent learning ability and generalization of SVM even with small samples, combined the RSM with  LS-SVM, then a new RSM was established based on LS-SVM, and it was used in model updating and damage identification. A weight method was proposed in kernel function, and the linear-and-Gaussian kernel function was proposed which combined the linear fitting ability of the linear polynomial kernel function and the nonlinear fitting ability of the Gaussian kernel function. At the same time, the training samples had a scale transformation, and the selection method of training samples was improved. By a identification example and experimentation, compared with the traditional sensitivity method, the results show that the improved RSM is better, and the convergence is also improved greatly.

Key words: response surface method(RSM), model updating, least square support vector machine(LS-SVM), kernel function, damage identification