Abstract: This paper mainly studies the application of deep learning and transfer learning in fault diagnosis of distribution network equipment. Deep learning models ( such as DNN, CNN, and DBN ) have significantly improved fault identification capabilities and feature analysis efficiency through hierarchical feature extraction and unsupervised pre-training. Transfer learning solves the problem of cross-scene and small sample learning through pre-training models and domain adaptive strategies. Multi-source data fusion and edge computing enhance the real-time and robustness of the system. However, there are some limitations in dealing with high impedance faults, interpretability of transfer learning, spatio-temporal asynchrony of multi-source data and communication delay.