One SVR modeling method based on kernel space feature

Abstract

Listen

Translate article

Support vector regression (SVR) has unique advantages in system modeling because of the structural risk minimization principle. However, its model generalization ability is subject to excitation signals and kernel parameters. An SVR modeling method excited by sinusoidal signals based on kernel space features is proposed to improve the generalization ability. First, the effect of amplitude of the sinusoidal excitation signals on the generalization ability of SVR model is analyzed. The following conclusions are drawn. The amplitude of the best excitation signal should be in the same order of magnitude as that of actual input signal and not smaller than the amplitude of the actual input signal. Second, based on the considered SVR modeling problem in the classification problem, the kernel space feature named inter‐cluster distance (ICD) is employed to determine a more effective searching range of the kernel parameter. Simulation experiments results show less training time and the superior generalization ability of the proposed method. Thus, the proposed method can be used as a guide to select the sinusoidal excitation signals and kernels parameters of SVR models. © 2017 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.