Review of Series Arc Fault Detection Methods in Photovoltaic Systems Based on AI

The series arc fault detection technology of photovoltaic system is an important research direction to ensure the safety of new energy power. This paper systematically reviews the research progress in this field at home and abroad, and analyzes the arc physical modeling, experimental platform construction, feature extraction method and intelligent detection algorithm at four levels. Regarding the experimental verification system, the existing research generally follows the UL1699B standard to build the basic platform. In the feature extraction method, timefrequency domain joint analysis significantly improves the feature differentiation through multi-scale signal processing, but faces the contradiction between high-frequency noise sensitivity and high computational complexity. As for intelligent detection algorithms, hybrid models based on deep learning can reach more than $99 \%$ detection accuracy in the laboratory environment, but in the actual power plant scenarios affected by component aging, environmental interference, and other factors, the performance degradation is significant. Future research needs to focus on the deep integration of arc physical mechanism and data-driven models, develop lightweight edge computing architectures, and build intelligent detection systems with strong crossscene generalization capabilities.