LSSVM parameters optimizing and non-linear system prediction based on cross validation

  • Authors:

  • Weimin Zhang;Chunxiang Li;Biliang Zhong

  • Affiliations:

  • Department of Computer Science and Information Technology, Guangzhou Maritime College, Guangzhou, China;Department of Computer Science and Information Technology, Guangzhou Maritime College, Guangzhou, China;Department of Computer Science and Information Technology, Guangzhou Maritime College, Guangzhou, China

  • Venue:
  • ICNC'09 Proceedings of the 5th international conference on Natural computation
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

With kernel function of radial basis function (RBF), least squares support vector machines (LSSVM) is used for non-linear system prediction in this paper. For limitation of gridding search method of cross validation, the parameters optimizing method is proposed to determine the regularization parameter and the kernel width parameter of LSSVM. And the methodology steps of this method are presented in detail. Compared with gridding search method, the applicability is validated through simulation experiment. In addition to higher generalization performance, the prediction results of nonlinear system show that this method can achieve higher prediction precision and cost less modeling time than BPNN.