Fault Diagnosis of Transformer Based on Probabilistic Neural Network

In order to improve the correct rate of transformer fault diagnosis based on three-ratio method of traditional dissolved gas analysis (DGA), a novel intelligent transformer fault diagnosis method based on both DGA and probabilistic neural network (PNN) was proposed. In this fault diagnosis method, it takes three characteristic values of the improved three-ratio method as its inputs and five transformer fault types as its outputs. And it selects the radial basis function, applies the one-against-one multiclass algorithm, and fully uses the superiority of PNN in processing finite samples. The efficiency of the proposed diagnosis method was tested by simulation of transformer fault diagnosis. The simulation results have shown that the better convergent speed, better generalization ability and higher accuracy are expressed in this proposed diagnosis method if a small data set is available.