The Taguchi- Artificial Neural Network approach for the detection of incipient faults in oil-filled power transformer

Abstract

This paper presents hybrid Taguchi-Artificial Neural Network to detect incipient faults in oil-immersed power transformer. It involved the development of Artificial Neural Network (ANN) designs and embedding Taguchi methodology to fine tune the parameters of a backpropagation feed-forward ANN. Detection of incipient faults in power transformer is essential because it is one of the fundamental equipments in the power system. Dissolved gas analysis technique was used as it has been found as a reliable technique to detect incipient faults as it provides wealth of information in analyzing transformer condition. This study is based on IEC 60599 (2007) standard and historical data were used in the training and testing processes. Comparative studies were conducted between heuristic ANN design and optimized hybrid Taguchi-Neural Network. The results show the effectiveness of the optimized neural network using Taguchi methodology.