Machine learning techniques for fault detection in simulated synchrophasors data 25kV distribution feeder

This paper introduces a groundbreaking approach to fault classification in power distribution systems, leveraging cutting-edge data mining techniques and pattern recognition algorithms. Utilizing two advanced machine learning algorithms, the study effectively identifies phase-to-ground faults in a 25 kV distribution feeder, notable for its high load and substantial penetration of Distributed Generation (DG). The test feeder was meticulously modeled using the OPAL-RT HYPERSIM simulation platform, which facilitated the generation of synchronized measurements adhering to the IEEE C37.118 protocol. Phasor Measurement Units (PMUs) were strategically installed at nodes 10 and 14 to capture high-resolution voltage and current phasor data. The research applies Support Vector Machine (SVM) and K-Nearest Neighbors (KNN) algorithms for an in-depth analysis and classification of faults. It rigorously evaluates the performance of these machine learning techniques in detecting faults within the distribution network and compares their effectiveness with traditional protective schemes. The study highlights the potential of machine learning to enhance fault detection accuracy and offers valuable insights into integrating these advanced methods into existing protection frameworks, aiming to improve reliability and operational efficiency in modern power distribution systems.