A Dimension Reduction Method and its Application in GIS Partial Discharge Pattern Recognition Research

Abstract

GIS partial discharge pattern recognition is an important part of its state evaluation, authors have set up 252kV GIS partial discharge detection simulation experiment platform based on UHF detection method, and designed four kinds of typical partial discharge models in laboratory, then established corresponding UHF signal mapping database through the experimental method, and also extracted the original feature parameters; because the original characteristic dimension is high, which is bad for pattern recognition, based on this, the article uses a species mean kernel principal component analysis method, it mapped the partial discharge original data samples to high-dimensional feature space, at first, it calculate all kinds of class mean vector data, and then do principal component analysis based on class mean vector space, build the class average kernel matrix, at last, the class kernel mean principal component analysis algorithm is established. Results show that characteristic of this method contained all the information of the original data, and dimension is less than GIS insulation defect category numbers, and it can realize data dimension reduction without information loss, which improve the pattern recognition rate.