A classification of partial discharge on high voltage equipment with multinomial logistic regression

Abstract

This document proposes a statistical approach in multinomial logistic regression to classify PD patterns into four categories listed as corona at high voltage side in air, corona at low voltage side in air, surface in air, and internal discharge. The independent variables in this multinomial logistic regression model are skewness, kurtosis, asymmetry, and cross correlation following the ø -q -n PD patterns obtained from the fingerprint analysis which is a digital signal processing technique for PD measurement. The experiments were set to simulate all four PD patterns to obtain statistical parameters into 10 independent variables from the fingerprint analysis. This document also applied stepwise model selection technique to reduce from 10 independent variables to 2 independent variables that not only reduces the complexity of the model estimated but also retains the accuracy of this predictive model to 96.2 percent. © 2006 IEEE.