Abstract: When the inverter of photovoltaic system is started, internal control algorithms such as maximum power point tracking (MPPT) will cause current transient changes, interfere with the correct identification of fault characteristics by DC series arc fault diagnosis device, and cause misoperation. Aiming at the problem that the arc fault detection device is prone to misoperation when the inverter is started, we proposed a fault detection method based on dimensionless feature quantity and grey correlation degree. Firstly, the RLC equivalent oscillation model of arc fault is analyzed. It can be seen that the arc current signal has a wide frequency band in the frequency domain. Then, the frequency domain characteristics of the measured current under arc fault and inverter working conditions are compared. It is found that there are differences in peak steepness and position between the spectrum diagram of fault and normal working conditions in the range of 1~ 20 kHz and 40~60 kHz. The spectrum features are extracted by using steepness, skewness, peak factor, impact factor, margin factor and waveform factor, and the fault is identified according to the grey correlation degree. Finally, the simulation platform and the actual photovoltaic system are used for experimental verification. The results show that the method can be adopted to effectively avoid the interference caused by inverter startup and improve the accuracy of fault identification.