Abstract: Fast and accurate fault location is very important to the running quality and reliability of power distribution system. At present, most of the fault location methods of multi-branch distribution are realized by adding measuring devices in distribution network. However, the topology structure of the distribution network is complex, the number of existing monitoring equipment is limited, and the large-scale installation of additional measuring equipment is expensive, which makes it difficult to be widely used in the actual system. Therefore, a fault location method for multi-branch distribution lines based on limited measurement information is proposed. Firstly, the nonlinear relationship between the capacitance information and the fault distance in multi-branch distribution lines is analyzed theoretically, and the feasibility of using deep learning to construct mapping function to complete the fault location task is proved. Then, the stack auto-encoder and long-short term memory network are used to establish an intelligent location model to reduce the error estimation of multi-branch distribution lines for fault location. Secondly, based on the measurement information in the distribution automation system, the determination and location for fault lines are realized by logical reasoning and intelligent location model. Finally, an intelligent localization implementation scheme based on deep transfer learning is proposed to improve the performance of the proposed method in various applications. The proposed location method is tested and verified on MATLAB/SIMULINK platform, and the simulation results show the effectiveness and generalization ability of the proposed method under complex working conditions and distributed generation access conditions. The results of this study can provide an auxiliary decision function for existing fault location methods.