To recognize the series arc fault in low voltage distribution network accurately, this paper proposes a method of series fault arc recognition based on an all-phase spectrum and deep learning. First, it deduces the generation mechanism of all-phase spectrum characteristics of load distortion signal in theory, and extracts the all-phase spectrum feature of a linear and non-linear load by an all-phase discrete Fourier transform. Secondly, it constructs a deep learning neural network model based on Logistic regression, and carries out deep learning training of all-phase spectrum characteristics under different loads and operating conditions. Finally, the sampling data of the fault arc experiment platform is analyzed. The results can accurately recognize whether a series fault arc occurs in a low voltage distribution network and distinguish the type of fault load. Experimental results verify the effectiveness of the proposed method. With the application of deep learning theory in the intelligent power system, the method is expected to be further studied and popularized. This work is supported by Major Project of National Natural Science Foundation of China (No. 51437003).