Abstract: According to the U-I output characteristics of PV array under different shades,this paper analyzes the generation mechanism of the DC series fault arc in PV system,and analyzes the characteristics of DC series fault arc signal by building a PV system fault arc test platform,and then introduces a method to detect DC arc fault under shading based on Volterra series. The proposed method firstly reconstructs the phase space of the current signal,and then establishes a Volterra series model in the reconstructed phase space to extract the time-domain kernel features,and uses the Kernel Limit Learning Machine（KELM）optimized by the lion colony algorithm（LSO）to identify fault arc. The experimental results show that the proposed method can accurately detect strong DC fault arcs in normal state and shadow state of PV system,and can also detect weak DC arcs.