A Machine Learning Based Hybrid Algorithm for Partial Discharge Localization in Power Transformers

Abstract

Locating partial discharges (PDs) in power transformers is crucial for preventing catastrophic damage. In this article, we first present a theoretical framework that proves the feasibility of determining a PD current uniquely by recording the tangential electromagnetic fields on its surrounding surface. This proof provides the analytical background required for the PD localization inverse problem by employing a machine learning approach. Three ML methods are combined in three steps of the proposed algorithm to achieve an accurate, yet near real-time identification of high-risk PDs. The effectiveness of the proposed method is demonstrated through simulations and experimental results, highlighting its potential for enhancing the reliability and safety of power transformers.