Arcing Fault is an important cause of electric fire. In order to solve the issues of unwanted tripping or rejected action that the existing arc fault protection methods have, a series arc fault detection method based on current similarity and high frequency energy is proposed. According to GB/T 31143, an arc fault experiment platform was built to carry out fault arc experiments under linear and nonlinear loads and their combination. And the characteristics of arc current were analyzed from the perspectives of time domain, frequency domain and "periodic similarity". Then Db 5 wavelet was selected for arc current preprocessing, the difference of cosine similarity of adjacent periodic current low-frequency approximation coefficient was used as the low-frequency feature of arc current, while the wavelet energy in frequency range of 3125 Hz~6250 Hz was used as the high frequency feature of arc current. On this basis, a fault arc diagnosis algorithm based on threshold comparison is proposed. Experimental results show that the proposed method can accurately identify fault arcs under single load and combined load conditions, and its arc detection time meets relevant standards.