Kernel principal component analysis for power quality problem classification

Abstract

This paper proposes the application of kernel principal component analysis (KPCA) for power quality (PQ) problem classification. First, the features of PQ signal are extracted using wavelet-multiresolution analysis. Then, KPCA captures the dominant nonlinear properties of the extracted features by transforming to a high dimensional feature space. The dimension of extracted features produced by KPCA can be reduced without loss of information of the original features. Finally, support vector machines (SVMs) are used to classify the PQ problem using the dominant components of KPCA. Simulation results with six types of PQ problem demonstrate that the proposed KPCA-based SVMs provides the superior classification performance of PQ problem to the conventional SVMs.