Design and Application of Intelligent Scalable Automatic Fault Detector for Commercial Photovoltaic Systems

Fault detection (FD) in photovoltaic power systems (PVPS) is critical for maintaining system efficiency and reliability. This study leverages real-time time series monitoring data from over 80 rooftop PVPS, collecting measurements and analyzing maintenance logs. These datasets undergo re-organization, filtering, and quality assurance processes to create comprehensive training, testing, and validation sets, considering variations in the number of inverters. Various dimensional extraction methods are applied to evaluate the accuracy of existing dimension reduction techniques, alongside the proposed feature extraction method. Comparative analysis demonstrates that the proposed method outperforms current approaches in accuracy and efficiency. Additionally, the intelligent fault detection process, including diagnosis and post-processing, enables precise localization and characterization of fault properties, providing deeper insights for fault management in PVPS.