Abstract: Artificial intelligent model based on data driven, especially the convolutional neural network(CNN), has already achieved great performance in distribution network fault diagnosis. While CNN relies on massive data, and the performance of model will decrease severely due to the lack of data. We proposed an active distribution network fault line selection method based on domain adaptive transfer learning. Firstly, a CNN incorporated attention mechanism was built, and fault characteristics of transient zero sequence current in active distribution network were extracted. Then, domain adaptive transfer learning was adopted, and the distance between source domain and target domain was decrease by the maximum mean discrepancy function, to solve fault line selection with small samples. Finally, the proposed method was verified in active distribution network under different operating mode by Matlab/Simulink. Simulation results verify the proposed method can realize high accuracy and robust fault line selection in active distribution network with small samples.