Application of Combinatorial Probabilistic Neural Network in Fault Diagnosis of Power Transformer

Abstract

Probabilistic Neural Network (PNN) overcame the shortcomings of entrapment in local optimum, slow convergence rate which was in BP algorithm. With enough training samples, PNN obtained the optimal result of Bayesian rules. Because of the fast training rate, the training samples can be added into PNN at any time. So, PNN is fit to diagnose the fault of power transformer and has auto-adaptability. In order to improve the classification accuracy, the conception of combination is introduced into PNN. The fault diagnosis of power transformer is consisted of 4 Probability neural networks in this paper. PNN1 is used to classify the normal and fault. PNN2 is used to classify the heat fault and partial discharge (PD) fault. PNN3 is used to classify the general overheating fault and severe overheating fault. PNN4 is used to classify the partial discharge fault, and energy sparking or arcing fault. The example shows that the effect of combinatorial PNN is a good classifier in the fault diagnosis of power transformer. The combinatorial PNN has better diagnosis accuracy than BPNN and FUZZY algorithm.