Detection of Transformer Winding Faults Using Wavelet Analysis and Neural Network

Abstract: This paper investigates the application of wavelet transform as a preprocessor for neural networks (NN) in identifying internal turn-to-turn faults in transformer windings. The faulty and normal signals generated by numerical simulation of ElectroMagnetic Transient Program (EMTP) are preprocessed using discrete wavelet transform (DWT). The mean values of the wavelet coefficients are input into an improved back-propagation neural network and an Elman recurrent network. Cross-validation is used to select the suitable architecture of the networks. The networks, after training, can decide if the measured signal is faulty or normal. The simulation results of four cases: improved BP-NN with wavelet preprocessing, Elman network with wavelet preprocessing, improved BP-NN without preprocessing, and Elman network without preprocessing are compared and discussed