In order to solve the problem of automatic evaluation of dispatcher training, a distribution network fault training evaluation method based on isolation information matrix similarity is proposed. The isolation information matrix is constructed through active isolation range set, passive isolation range set, power loss load and load number, which is the information model of dispatcher and trainer fault simulation training. Based on the artificial intelligence reasoning theory, the similarity between the real-time and standard isolation information matrixes is described by similarity theory to realize the evaluation of dispatcher fault simulation training. Through the active isolation range longitudinal similarity, passive isolation range longitudinal similarity and important index longitudinal similarity, the similarity analysis of the single element of real-time and standard isolation information matrix is realized. Through the similarity of single element and Euclidean distance, the fault feeder horizontal similarity is realized, and the isolation information matrix overall similarity is realized by combining multi feeder fault and Hamming distance. The analysis shows that the method can better reflect the influence of dispatcher fault isolation and transfer operation on the distribution network operation, which can effectively distinguish the evaluation grade between different isolation operations and has good practical value. This work is supported by National Natural Science Foundation of China (No. 61403321).