Comparative Analysis of Fault Classification in Hybrid Microgrid Using Different Machine Learning Algorithms

The reliability of the power system is impacted by protection challenges in microgrids due to the increased integration of distributed generators that supply local loads. To enhance overall power system resilience, addressing these challenges is crucial. Conventional protection systems face challenges in reliably identifying bidirectional power flow during power exchanges between microgrids and the main grid. Adaptive intelligent protection methods offer a solution, adapting to grid changes, operational modes, and energy resource variations. This study conducted a comparative analysis of implementing different machine learning algorithms, such as Artificial Neural Networks, Convolutional Neural Networks, and k-Nearest Neighbors, for fault classification in a hybrid microgrid in standalone mode. The input dataset comprises three-phase currents and voltages, aimed at accurately identifying 11 fault types and instances without faults. Microgrid simulations are conducted in MATLAB/Simulink, while Python software is used for testing and training. This research assesses the effectiveness of these algorithms in improving fault classification within microgrids using accuracy as the primary parameter. The findings indicate that Convolutional Neural Networks outperform Artificial Neural Networks and k-Nearest Neighbors in this study.