Machine Learning Applications for Online Partial Discharge Detection, Classification, and Localization in Power Transformers: A Review

The power transformer is a crucial asset and a fundamental component of the power grid. Assets undergo aging due to the stresses present in insulation materials. Partial discharges (PDs) are the most common fault source in power transformers and an excellent indicator of aging. The detection, classification, and localization of PD activities in power transformers are persisting challenges, while techniques utilizing machine learning (ML) are widely sought to deal with those challenges. Existing ML techniques show promising results with an elevated level of accuracy and precision. However, there is a lack of conventional ML-based real-time monitoring capability. Therefore, this paper presents a comprehensive review of the application of ML techniques for online PD activity detection, classification, and localization in power transformers, focusing on supervised, unsupervised, semi-supervised, and reinforcement learning techniques. In addition, this paper explores the challenges, future trends, perspectives, and outlook of machine learning for online transformer fault analysis.