Assessing the Degradation of Transformer Oil From Partial Discharge Measurement Data Using Histogram Similarity Measures
Lakshmi Tharamal, P. Preetha, T. K. Sindhu
Abstract
In this paper, a method is proposed to classify the in-service transformer oil by assessing the level of degradation in oil. Initially, 135 oil samples are classified into Class 1, Class 2 and Class 3 as per IEEE C57.106 (2015) standard by the conventional techniques which require five tests. The proposed method attempts to classify the oil samples by a single non-destructive Partial Discharge (PD) test measurement. The PD data of the 135 transformer oil samples are represented as one-dimensional histograms and classified by statistical analysis using Histogram Similarity Measures including Cross-correlation test, Kolmogorov-Smirnov distances, and Chi-square test. This classification technique achieves an accuracy of 94.9 %. The results are further subjected to class likelihood measures in the post-processing stage and this improves the accuracy of classification to 97.5 % establishing the efficiency of the proposed method.