Modeling of dry type transformer winding temperature behavior using artificial neural network

Transformer is one of the most important components in terms of high cost and investment in transmission and distribution systems. Losses of core and windings cause to significant temperature rising in transformers. Due to effect of transformer insulation temperature on the transformer life expectancy, the transformer insulation temperature which accelerates the rate of aging of the insulation should be known. In this study, the Artificial Neural Network (ANN) models are presented to model the dry type transformers winding temperature behavior. The aim of using the neural network is its ability to learn complex and nonlinear structures. Because the behavior of transformer winding temperature has dynamic characteristic, Recurrent Neural Network models are used. The same network structure and training algorithms were applied to two different experimental data sets obtained from 5 kVA and 3 kVA transformers. As a result of evaluations of using performance determinant factor, Nonlinear Autoregressive with Exogenous Inputs (NARX) model trained with Bayesian Regularization algorithm was determined as the most suitable structure for our system.