Abstract: Aiming at the problem that it is difficult to locate faults quickly and accurately in intelligent substation communication networks, a fault diagnosis method based on graph filtering neural network is proposed. Starting from the intelligent substation communication network topology graph and taking different devices as nodes, the feature information shown under different detection nodes during fault state is analyzed and the feature expression of different nodes is proposed. The expansion of fault samples is realized by changing different component faults and network component configurations, thereby enabling the communication network to generate new operating states. The fault data is expressed in the form of graph data and combined with the theory of graph filtering neural network to build a fault diagnosis model. Finally, taking some intervals of 220 kV smart substation as an example, by comparing the fault localization effect of different methods, the proposed method is verified to have the advantages of lower sample requirement, higher fault localization accuracy and better fault tolerance.