Study on system identification based on Kernel function KPCA-SVR

In nonlinear systems, the structure identification is one of the difficulties, including the dimensionality selecting of sample space and the inside structure confirming of model, which impacts the accuracy and generalization ability of model. Aimed at that problem, a novel system identification approach based on KPCA (kernel principal component analysis) and SVR (support vector regression) is presented. Firstly, the nonlinear components of sample space are extracted by KPCA, which confirms the dimensionality of sample space. Further, in order to confirm optimal inside structure of model, SVR with structure risk minimization (SRM) is utilized to realize the optimal identification. Simulation results reveal that KPCA-SVR is effective in solving nonlinear system identification.