Leveraging artificial intelligence through neural network error residuals for advanced fault detection and classification in photovoltaic systems

Faults undetected in photovoltaic (PV) installations lead to wasted energy, system failures, and damage to solar modules. To mitigate these risks, intelligent and robust fault-detection algorithms are essential for PV solar energy systems. This study introduces an artificial intelligence (AI)-driven fault diagnosis technique utilizing artificial neural networks (ANNs) to detect and classify faults in standalone PV installations. In this work meteorological and electrical data were retrieved from a 200 W standalone PV system. The approach combines ANN-based regression and classification models for effective fault diagnosis. The regression model uses meteorological variables such as backplate temperature and solar irradiance to predict PV voltage and current. On the other hand, the classification model uses the residuals between predicted and measured values to identify the number of defective panels under partial shading and open-circuit fault scenarios. Trained on historical in-situ normal and faulty operating data, the proposed AI-based methodology achieves approximately 90.9% accuracy in electrical parameter forecasting and fault classification, demonstrating enhanced reliability and decision-making for PV system diagnostics.