Feature optimization selection and dimension reduction for partial discharge pattern recognition

Abstract

Hundreds of features have been extracted from phase resolved partial discharge (PRPD) pattern and PD waveforms to represent and recognize typical defects. Several feature selection and dimension reduction methods for pattern recognition are presented in this paper. Feature selection algorithms including forward feature selection, backward feature selection and floating forward feature selection (FFFS) are adopted to optimally select the features. |Four dimension reduction algorithms such as principal component analysis, linear discriminant analysis, kernel principal component analysis and generalized discriminant analysis (GDA) are used to further reduce the dimension of features. In order to compare the effectiveness of different selection and reduction techniques, PD tests on artificial PD defect models are performed. The results indicate that the FFFS and GDA are the optimal selection and reduction method, respectively.