Abstract: In this paper, a series arc experiment platform for photovoltaic system is built, 50 sets of arc data under different environments are collected, energy characteristics and entropy characteristics in frequency domain and wavelet domain are extracted, and the distribution of characteristics is drawn. The discrimination ability of 18 features is evaluated by introducing the discrimination degree. The evaluation result shows that the 10-30 kHz spectrum energy has the best discrimination ability to the arc, and the discrimination degree between the arc state and the normal state reaches 98.49%. The influence of inverter startup, light intensity change and different environment on characteristics is studied, and the anti-interference ability of 10-30 kHz spectrum energy in photovoltaic system is analyzed. A series arc detection device is made and its threshold setting and promotion method are proposed. The device can accurately detect and quickly extinguish the arc in many experiments, and there is no misoperation in the immunity experiment. The action time can reach 41 ms.