Exploring Machine Learning Techniques for Fault Classification in Wind Farm Collectors

The massive increase in integration of Inverter-Based Resources (IBRs) into electrical power systems increases system complexity and affects power quality due to the unconventional fault contributions of IBRs. Therefore, this work addresses the complexity attributed to IBRs in fault classification in wind farms. A test system was modeled in PSCAD/EMTDC software and simulating various fault scenarios within a wind farm. The generated data were used as input of intelligent classification methods. The study explores the potential of Machine Learning (ML) methods through the WEKA software (Waikato Environment for Knowledge Analysis). The results demonstrated the effectiveness of ML methods for fault classification, with accuracies exceeding 99%. Furthermore, through the analysis of noise impact on the signal, power generation levels, and generalization, the robustness of the proposed methodology was proven, as the reduction in its accuracy was less than 2.5% in the most critical scenarios.