Transformer Fault Diagnosis Based on Elman Network

Power transformer fault is characterized by diversity, irregularity and uncertainty. So Elman Network fault diagnosis method is proposed, for the effective realization of transformer fault diagnosis. Key elements are selected from the transformer gas content acquisition parameters as parameters for improved three-ratio method, which is the input of Elman network. Four kinds of fault types are firstly selected, which constitute five kinds of network output parameters with the normal working condition. A different number of hidden layers are set in the network, and the simulation results of different hidden layers are obtained through the training and test of the network. By contrast, the fault type of transformer can be effectively judged, and the conclusion can be made that the number of the errors is the smallest when there are 8 hidden layer nodes, with the accuracy rate up to 87.5%. If the judged transformer fault types need to be increased, we can just increase the number of the network output nodes, indicating the establishment of Elman network for transformer fault diagnosis can undertake transformer fault diagnosis well, with the adjustability, which can meet the judgment request from different quantities of judge transformer fault types.