A Comparative Analysis of Different Machine Learning Algorithms for Classification of Partial Discharge Signals under HVDC

The identification of Partial Discharge (PD) sources in an insulation system is of great practical significancy, as one can estimate the number of voids present in the insulation from it. However, such identification become an uphill task when the operating field is high voltage DC (HVDC) primarily because of absence of phase resolved data which is available in case of AC. Therefore, for classification of PD sources under HVDC, the help of machine learning algorithms can be sough of. In present work, a comparative performance of four well know machine learning algorithms i.e. Random Forest, KNN, SVM and Gaussian Naive Bayes (GNB) is analyzed for the purpose of classification of PD signals coming out from different kind of defect under HVDC. Three types of defects i.e. Single Void, Serial Void and Parallel Void have been considered for the comparative analysis. It was observed that Random Forest Algorithm gives best performance among all of them.