Fault Detection and Localization in Photovoltaic Systems Using Extreme Learning Machine: Handling Short Circuits and Partial Shading

Abstract

Reliable photovoltaic (PV) system operation is vital for sustainable solar deployment. This study proposes a fast fault detection and localization method using Extreme Learning Machine (ELM) for short circuits, partial shading, and combined faults in PV arrays. A simulation-based dataset spanning diverse faults and load profiles addresses the scarcity of labeled field data. The model was assessed via confusion matrices and classification metrics: accuracy, precision, recall, and F1-score. Results show 76% overall accuracy and 81% average precision, indicating reliable separation of distinct fault classes. However, overlapping signatures confused combined faults, with precision dropping to $\mathbf{4 2. 1 3 \%}$ in specific classes. Planned improvements include ELM parameter tuning, richer features incorporating temporal and frequency-domain attributes, and broader training data. Integration with SCADA and embedded platforms is envisioned for deployment to enable real-time monitoring. Overall, the approach advances intelligent PV system health monitoring, enhancing reliability while reducing maintenance burden and operational costs.