Design of a Robust and Fast Fault Diagnosis Strategy of a Stand-Alone LVDCMG Using Gaussian SVM & Weighted KNN Algorithms

Despite the numerous advantages offered by DC microgrids, ensuring their protection remains a complex task. Protecting these systems poses a significant challenge that hinders the full realization of the benefits and applications of DC microgrids. Limited expertise in managing DC faults and a lack of DC circuit breakers (DCCBs) with higher capacities have hindered the progress of protection measures in DC microgrid systems. The focus of the work is to address the issue of fault identification and categorization in low-voltage DC microgrids (LVDCMGs). Further, a two-layered machine learning approach based on Support Vector Machines (SVM) and weighted K-nearest neighbor (WKNN) algorithms have been proposed. The proposed technique is implemented using MATLAB to simulate a 96V LVDCMG with photovoltaic (PV) panels, battery energy sources, and loads. Local voltage and current measurements are identified as the most informative attributes for fault diagnosis using the statistical method of feature selection and are utilized as training samples for the classifier models. The performance of various kernel functions of SVM classifiers has been analyzed. The simulation results reveal that the medium Gaussian SVM classifier outperforms compared to the other variants of SVM. The WKNN classifier classifies the faults as either PG or PP to enable fast appropriate corrective actions. The results demonstrate that the proposed two-layered protection scheme achieves 100% accuracy in fault detection. The performance of the classifier models is further validated using random data samples from the network.