Hybrid multilayer perceptron neural network for transformer health index monitoring

Dissolve Gas Analysis (DGA) for transformers is used to differentiate between a transformer in good condition or the one which needs to schedule for maintenance. The main goal of DGA is to identify more precisely problems caused by the various gas formations in the transformer and encountered. Key Gas Method (KGM) analysis is one of the DGA's techniques often used. KGM is used in forecasting the health index of the transformer based on formational of gases in transformer. In the research, several classifiers are arranged to obtain the best performance based on four (4) configuration factors. The hybrid multilayer perceptron (HMLP) network, multilayer perceptron (MIP) network, K-Nearest Neighborhood (KNN), Linear Discriminant Analysis (LDA) and Support Vector Machine (SVM) are used to do the classification. The HMLP network outperform other classifier with 92.35% on accuracy and 0.78 on the MSE, respectively. Three different training algorithms selected to train HMLP with Backpropagation (BP), Lavenberg Marquardt (LM) and Bayesian Regularization (BR) training algorithms. At the end simulation, BR training algorithm shows the best performance with accuracy of 94.13% and 0.39 on MSE, respectively.