Transformer fault diagnosis based on neural network and fuzzy theory

Transformer fault diagnosis method was studied based on neural network by combined method of fuzzy theory and neural network. Three ratios, w(C2H2)/w(C2H4), w(CH4)/w(H2) and w(C2H4)/w(C2H6) were calculated by transformer fault characteristic gas data, fuzzy membership function was utilized to process three-ratio data. Then, fuzzy processed three-ratio data was input to the neural network, fault type coding was the neural network output, BP neural network was established and trained. Finally, the trained radial basis function neural network was utilized to diagnose transformer fault diagnoses, such as low energy discharge, high energy discharge, partial discharge, low/high/middle temperature overheating. Fault diagnosis experimental results show that diagnostic accuracy is upto 85.4%.