Abstract: EDGs are important equipment related to the safety of nuclear power plant. The timely and accurate fault diagnosis to ensure the normal operation of EDGs is of grea t significance to nuclear safety. In this paper, the fault types of nuclear EDGs are sorted out, the fault sets, parameter sets and fault characteristic data are summarized, and the SOM neural network is established by using the MATLAB neural network toolbox to realize the clustering and simulation of the fault types of EDGs, so as to accurately diagnose the single and double faults of EDGs at certain time. The simulation results show that the SOM neural network have fault identification ability, but new fault samples are needed to build up on two or more fault diagnosis process, which are more complicated, and does not have actual operation. Then the fault diagnosis of compound neural network is studied. The accuracy of SOM neural network diagnosis is maintained while the operability is taken into account.