Abstract: DC arc faults often result in fire accidents, therefore, accurately identifying arc faults is crucial for ensuring the safe operation of photovoltaic systems. Traditional fault detection methods face limitations in handling the variability of arc characteristics, dependence on feature extraction, and real-time processing. To address these issues, this paper proposes an arc fault detection algorithm based on the ArcNet architecture. This method combines multimodal data, including current, voltage, and solar irradiance, to provide efficient and accurate fault identification. Data from a photovoltaic system fault testing platform, including both normal and arc fault data, was collected. The key characteristics of arc faults were identified, and the ArcNet algorithm was designed based on these findings. Experimental results show that the overall accuracy of the ArcNet algorithm on the test set is 98.05%, with a false alarm rate of 1.79% for normal data and an undetected rate of 2.59% for arc fault data. Compared to traditional algorithms, ArcNet shows significant advantages in terms of accuracy, false alarm rate, and undetected rate, which will provide effective technical support for the safe operation of photovoltaic systems.

Key words: photovoltaic system, DC arc, fault characteristics, ArcNet, arc fault