Enhancing Power Grid Reliability: Evaluating Machine Learning Algorithms for Fault Classification in Inverter-Based Generators Interconnection Lines

Through the transition of the energy matrix, which has resulted in a more significant share of renewable energy sources, there is a massive integration of Inverter-Based Resources (IBRs) in Electrical Power Systems (EPS). This integration increases system complexity and affects power quality due to the unconventional fault contributions of IBRs. Therefore, in recent years, several researchers have studied the challenges of IBRs for conventional protection systems associated with EPS. This work addresses the complexity attributed to IBRs for the fault (short circuits) classification task in transmission lines connected to a wind farm. The study explores the potential of intelligent methods using machine learning algorithms based on decision trees and association rules through the WEKA (Waikato Environment for Knowledge Analysis) software. The results demonstrated the effectiveness of intelligent methods for classifying faults, with success rates exceeding 99%. However, it was found that the signal noise level significantly impacts the accuracy of the evaluated methods. Furthermore, some intelligent techniques have demonstrated the advantage of visibility into their decision-making, returning trees or rules with low complexity after training, which would facilitate the interpretation of this process by end users and its practical implementation.