Comparison of Machine Learning Algorithms for Classification of Partial Discharge Signals in Medium Voltage Components

Abstract

Partial discharge (PD) diagnosis is an effective tool to track the condition of electrical insulation in the medium voltage (MV) power components. Machine Learning Algorithms (MLAs) promote automated diagnosis solutions for large scale and reliable maintenance strategy. This paper aims to investigate the performance of two MLAs: Support Vector Machine (SVM) and K-Nearest Neighbors (KNN) for the classification of different types of PD sources. Suitable features are extracted by applying statistical parameters on the coefficients of discrete wavelet transform (DWT) for observing the performance of both MLAs. The performance of the algorithms is evaluated using key performance indicators (KPIs); accuracy, prediction speed and training time. Besides KPIs, a confusion matrix is presented to highlight the accurately classified and misclassified PD signals for the SVM algorithm. Comparative study of both algorithms demonstrates that SVM provides better results as compared to the KNN algorithm. The proposed solution can be valuable for the development of automated classification.