Application of support vector machines to nonlinear system identification

Abstract

It is a key research issue for support vector machines (SVMs) to choose kernel function for approximating a function. Different kernel function forms different SVM model that has distinct performances. In this paper, after the nonlinear system identification method using SVM is discussed, the criterion of choosing kernel function for system identification is given, and the effect of parameters are discussed. In the experiment, several kernel functions are used to form different SVM models that are used to identify a typically nonlinear system, respectively. To analyze the effect of a parameter on SVM, plenty of parameters are employed to make the system identification experiment. A large number of experimental results show that radial basis kernel function is a good choice for identifying a nonlinear system using SVM.