AC Fault Detection in On-Grid Photovoltaic Systems by Machine Learning Techniques

The increasing integration of solar energy into the power grid necessitates robust fault detection and diagnosis (FDD) guidelines to ensure energy continuity and optimize the performance of grid-connected photovoltaic (GCPV) systems. This research addresses a gap in the literature by systematically evaluating machine learning (ML) algorithms for the detection and classification of AC-side faults (inverter and grid faults) in GCPV systems. We utilized three commonly employed algorithms, namely K-Nearest Neighbors (KNN), Logistic Regression (LR), and Artificial Neural Networks (ANNs), to develop fault detection models. These models were trained using a monthly electrical dataset obtained from the AYCEM-GES-GCPV power plant in Giresun, Turkiye, and their performance was rigorously evaluated using classification accuracy, Area Under the Curve (AUC), and Receiver Operating Characteristic (ROC) analyses. The results demonstrate that the algorithms are highly effective in fault detection, with AUC values consistently exceeding the critical threshold. The obtained accuracies for KNN, LR, and ANN were 0.9826, 0.782, and 0.7096, respectively. These findings emphasize the high effectiveness of ML algorithms, with KNN exhibiting the best performance, for identifying AC-side faults in GCPV installations. While the study focused on AC-side fault detection, subsequent work developed a smart card module to identify complex DC side electrical faults and built a PV array for experimental testing.