Abstract: The series arc in the photovoltaic DC system cannot be detected and cut off by the protection device, the arc continue to burn and generate high temperatures, which cause great harm to the safe operation of the photovoltaic system. Aiming at which, Cassie arc model is analyzed, the electrode spacing is introduced into the model, and the relationship between electrode spacing and arc stable combustion is analyzed. Then, the photovoltaic series arc simulation model is established, and the current data in different regions during steady combustion is obtained, feature vectors are determined by analyzing the arc fault characteristics, a probabilistic neural network model is built to identify the arc fault, and BP neural network model is established as comparison, the series arc category corresponds to the area to realize the identification of the series arc fault area. The results show that the improved Cassie arc model can be used to characterize the dynamic process and stable combustion state in different areas; under the same data, the recognition rate of the probabilistic neural network model is higher than the BP neural network, and the series arc fault area can be accurately located.

Key words: photovoltaic system, series arc, Cassie arc model, electrical fire, probabilistic neural network