Abstract: With the high penetration of distributed generators(DGs)，the radial structure of conventional distribution system w illchange to multi-terminal type．Influenced by the short-circuit current injected from distributed generators，the traditional distribution network fault line selection method may fail． So choose the right line selection method, find the fault line and remove the fault, the safe operation of the grid has a very important significance. Firstly, an accurate active distribution network model is established in MATLAB environment. In this model, single-phase ground fault that may occur is considered. Then, the data collected from the output characteristic of the converter under the condition of single- And train it in TensorFlow environment. Finally, the trained model is used to detect the state of single-phase ground fault so as to achieve the effect of fault line selection. The method mainly collects the characteristic quantity output by the converter when the single-phase grounding fault of each line of the system is set up, and automatically excavates the logic relation behind the characteristic data so as to improve the redundancy of the occurrence of the fault threshold, and As the number of acquisitions increases, the accuracy of fault line selection will be improved. The results show that this method effectively improves the accuracy of single-phase ground fault line selection.