Efficient Fault Classification in Distributed Generation Systems using M-KNN and Grid Search Techniques

To meet the growing demand for electricity and ensure a sustainable future, there is a significant shift towards distributed generation (DG), including non-renewable energy sources. The stability and reliable operation of DGs, which involve non-uniform power generation, present challenging problems. Proper fault diagnosis and mitigation are crucial in these systems. Consequently, reliable fault identification and mitigation are essential to ensure the trustworthiness and functionality of DGs. Established mathematical models for fault identification, location, and system isolation can be time-consuming and inaccurate. With advancements in machine learning (ML) and artificial intelligence (AI), these technologies have found applications in distributed generation systems (DGS). Therefore, this article investigates the use of Mutual K-Nearest Neighbors (M-KNN) for fault identification. This study considers a 100 km grid-connected distributed generation system comprising two distributed generators (DGs), simulated using MATLAB® to obtain data for ten different faults at various locations spaced at 2 km intervals. Subsequently, M-KNN in Python is employed for fault classification to accurately determine the nature of faults. To enhance the robustness of the model, a grid search approach with and without cross-validation (CV) is utilized. The achieved training and testing accuracies approach 99%, surpassing the performance of Basic KNN (B-KNN) models.