SVM和基于PCA、PLS的SVM在非线性辨识中的比较研究

Comparison of SVM,PLS-based SVM,and PCA-based SVM for Nonlinear System Identification

对于同一个非线性系统,比较单纯ε-不灵敏支持向量机方法和基于主元提取、基于偏最小二乘提取的ε-不灵敏支持向量机方法在输入相关和不相关两种情况下的拟合性能和抗干扰性能。仿真结果表明:当输入变量之间存在相关性时,基于特征提取的方法优于直接采用ε-不灵敏支持向量机的方法。

To compare the performances of e-insensitive SVM, PLS-based SVM and PCA-based SVM, two cases of the same nonlinear system for identification were considered. One is that the input variables are correlated, the other is they are uneorrelated. The results showed that PLS-based SVM and PCA-based SVM perform much better than that without feature extraction in the case of variables correlation.