Improvement to reduce training time of back-propagation neural networks for discrimination between external short circuit and internal winding fault

Abstract

This paper proposes the improvement technique to reduce training time of back-propagation neural network. The decision algorithm based on the hybrid of discrete wavelet transform (DWT) and back-propagation neural network (BPNN) has been proposed to classify between external fault and internal fault in power transformer. The DWT is employed to decompose high frequency component of post-fault differential current signals and used as an input pattern for the training process of a neural network in a decision algorithm with a use of the BPNN. The proposed technique is compared with conventional training process of BPNN in terms of average accuracy and training time process. The obtained results show that the proposed technique can reduce of training process duration time and is very effective in classifying between external fault and internal fault in power transformer with satisfactory accuracy.