A DC arc detection method for photovoltaic (PV) systems

PV arc-faults can cause fires, damage property, and endanger people's lives. This paper proposes a method for detecting DC arcs using artificial intelligence (AI). The four steps for arc detection are thoroughly described. After removing the low-frequency range (41 kHz) and high-frequency range (>102.5 kHz) components, the middle frequency range is left for arc analysis. For AI analysis, eight inputs are used. The time sequence for the tasks is also explained, where the parallel task configuration is adopted to save the time. Furthermore, AI model training for arc detection is described, including both offline and online training. In addition, three different types of arc detection system architectures are depicted. There are three layers in the architectures: the PV-end layer, the inverter-level layer, and the cloud layer. Depending on the architecture, the algorithm is located in different layers. Furthermore, the hardware of the arc detection system is explained, as is the self-testing circuit. Finally, an experimental platform is built, and experimental results are presented to validate the proposed method.