Abstract: Aiming at the problem that the existing protection methods for flexible DC distribution network are not sensitive to highresistance grounding faults, a high-sensitivity fault identification strategy based on deep belief networks（DBNs） is proposed. The strategy uses the feature extraction ability of DBN to extract the features of positive and negative transient voltage waveform data in time domain for fault identification. The simulation sample data is used to train the DBN, and the optimal parameters of the model are obtained to form a DBN model suitable for DC distribution network fault identification. The performance of the model is tested based on PSCAD/EMTDC simulation platform. The test results show that the proposed identification strategy has good sensitivity and generalization ability for high-resistance grounding faults.