Enhancing fault detection and classification in secondary distribution networks: A comparative analysis of bagging, boosting, decision tree and neural network approaches with non-contact magnetic sensors on distribution transformer /

Abstract

To ensure the reliability and uninterrupted flow of electric power, accurate and rapid fault detection and classification within distribution networks are crucial for protecting expensive equipment, such as distribution transformers. In this article, instead of the current and voltage signals measured by conventional current and potential transformers, non-contact magnetic sensors are placed optimally around the transformer tank to capture magnetic flux density (MFD) patterns during SC and OC fault conditions. The non-contact magnetic sensors offer safety, accuracy in fault detection, and a non-complex method, as changing current in transformers naturally generates a magnetic field in the surroundings. To achieve this objective, electromagnetic analysis of a distribution transformer is conducted using finite element analysis (FEA). The captured MFD patterns are then utilized for intelligent fault detection and classification by applying machine learning algorithms such as decision tree (DT), gradient boosting (GB), random forest (RF), and artificial neural network (ANN). The ANN model achieved an accuracy of 98.71% and 92.38% for both fault cases, slightly lower than the RF model’s accuracy of 99.74% and 93.02% for both fault cases in accurately detecting and classifying 18 fault case scenarios alongside the no-fault case scenario. The RF model performed its prediction for 7203 complex SC MFD case samples in just 0.0759 seconds and for 4802 complex OC MFD case samples in just 0.0413, proving it as one of the fastest algorithms for fault detection and classification.