Abstract: Aimed at the non-stationarity and particularity of the vibration signals of hydroelectric sets, a new fault diagnosis method combining singular spectrum entropy based on ensemble empirical mode decomposition (EEMD) with a self-organizing feature map network (SOM) is presented. First, EEMD was used to decompose the vibration signals of a hydroelectric unit to obtain their intrinsic mode function (IMF), and then singular spectrum decomposition was performed to obtain their singular spectrum entropy: a dynamic eigenvector that characters the signals. Finally, this feature vector was input into a trained SOM neural network for automatic recognition of the fault. The results show that this method can extract the fault characteristics of the unit accurately and it has a higher recognition accuracy and faster calculation speed.