Distribution network is the place with the most short circuit faults and low degree of intelligence. At present, the zero-sequence current amplitude comparison method is mainly used for fault diagnosis of grounding fault, but there are some problems such as low recognition rate of non-grounding fault and unable to identify quickly. 5G communication technology is adopted, three-sequence composite current detection method is proposed, and generalized regression neural network (GRNN) is combined to realize real-time transmission and rapid decision-making of fault diagnosis in distribution network. The test results show that the recognition rate of non-grounding fault can be improved by more than 20%, which solves the problems of unreliable communication and low intelligent fault diagnosis.