Abstract: Knowledge graph associates a large number of semi-structured/unstructured text data in the distribution network to improve the efficiency of the distribution network fault handling. However, it is difficult to use the multi-source heterogeneous text data in the distribution network for the deep learning model training, and there is a high labeling cost of the text data in the power field. In this paper, the pre-training method is used to build a deep learning model to identify the named entity of the fault handling data. The knowledge graph technology is used to store and apply the data so as to assist the regulators in making fault handling decisions. Firstly, taking the distribution network equipment account data, the fault handling data, the dispatching regulation data and the distribution network defect data as the objects, the framework and method of building the knowledge graph for the distribution network fault handling are proposed; Then, aiming at the problem of insufficient data available for the deep learning model training in the distribution network, the entity recognition model is constructed by using the pre-training method to extract the domain unstructured knowledge of the distribution network; Next, the design experiment proves the effectiveness of the model constructed in this paper with the F1_score of the model as 86.3% and the accuracy as 95.16%; Finally, the Neo4j graph database is adopted for highly visual management of the knowledge graph. The application process of the distribution network fault handling knowledge graph is given, which can effectively improve the decision-making efficiency and handling effect of the distribution network regulators.