As the scale of the distribution network continues to expand, the hazard caused by the occurrence of single-phase ground faults is also significantly increased. In order to avoid further escalation of the fault, measures must be taken quickly to remove the fault. Distribution network fault identification is helpful to quickly identify the cause of the fault and take corresponding measures to remove the fault. At the same time, fault identification is also a prerequisite for fault line selection. In view of the above situation, this paper introduces a new method for fault identification using deep neural networks. The results show that the method can identify various single-phase ground faults of small current grounding systems and the identification accuracy is high, and the identification accuracy is less affected by noise pollution than traditional artificial neural networks.