Fault Detection and Classification in the Photovoltaic Arrays using Machine Learning

Detection and classification of photovoltaic (PV) array faults are crucial for increasing grid susceptibility and decreasing power losses. Different PV array faults, such as Line-to-Line, Line-to-Ground, Partial shading, and Complete shading, make fault detection and classification challenging. The objective of this research is to propose a precise and rapid fault detection and classification technique based on the application and comparison of twelve supervised machine learning classifiers. Statistical analyses were utilized to construct the features matrix of the PV output current. Under challenging environmental and technical conditions, PV arrays with LL and LG faults were examined, and partial shading faults with single, double, and triple-shaded modules were distinguished from complete PV shading. After applying three cross-validation strategies, Extra Trees classifier showed the best performance among others, therefore, it was selected. Simulation results on unknown dataset demonstrated a 100% accuracy for fault detection and a 94.44% accuracy for fault classification using Random Forest Classifier (RFC).