A Comprehensive Review of AI-Driven DC Arc Fault Detection in Photovoltaic Systems

Photovoltaic (PV) systems now account for over 1.3% of global electricity generation and are expanding rapidly. DC arc faults pose severe safety risks among PV system faults, including fire hazards, equipment damage, and system failures. Traditional protection methods such as overcurrent devices and threshold-based detection often fail to distinguish arc faults from normal system noise reliably due to series arcs' intermittent and low-current nature. To address these problems, intelligent methods, including machine learning (ML), deep learning (DL), and hybrid approaches, have emerged as promising solutions offering superior accuracy, adaptability, and real-time performance. This paper presents a state-of-the-art intelligent approach for DC arc fault detection in PV systems. We explicitly compare ML and DL algorithms, highlighting trade-offs in computational efficiency, data requirements, and hardware constraints. Key implementation challenges, limited real-world datasets, and high computational costs for edge deployment are analyzed. Future directions focus on bridging gaps such as edge computing for real-time detection, synthetic data generation, and interpretable AI models. The findings aim to advance PV safety standards and enable scalable renewable energy integration.