Confidence-Level-Based Semi-Supervised Machine Learning Approach for Partial Discharge Signal Classification

Partial discharge (PD) classification is crucial for the condition monitoring of equipment with high power density. Recently, machine learning-based approaches have been developed for the detection and classification of PD signals. Most of these PD classification techniques are based on the availability of labeled training data. However, in practice, not all data may have labels. In this paper, we classify various types of PD signals using confidence-level-based semi-supervised learning. The algorithms investigated are Naïve Bayes, extreme gradient boosting (XGboost), and random forest. The performances of these algorithms are compared in terms of their PD source prediction accuracy. The results show that the performance of XGboost is much more superior than Naïve Bayes and random forest.