DRY TYPE TRANSFORMER WINDING THERMAL ANALYSIS USING DIFFERENT NEURAL NETWORK METHODS

16471

Artificial Neural Network (ANN) has become a modeling tool frequently used in applications and analyzing the complex problems in different disciplines. In this study, the thermal model of dry type transformer winding is modeled by using three different ANN structure and the most successful network structure is determined related to this modeling. The Neural Networks models used in this thermal modeling analysis are; i- Feed-forward Neural Networks, ii-Elman Recurrent Neural Networks and iii- Nonlinear autoregressive with exogenous inputs (NARX). In addition, Levenberg-Marquardt and Bayesian Regularization teaching algorithms are applied to the three ANN models and the results are compared. The same training algorithm and network structure are applied on two different experiment test data obtained from 3 kVA and 5kVA dry type transformers. The network structure and training algorithms are evaluated by using performance determinant factor. By comparing the results, it is determined that Bayesian regularization is the best training algorithm and NARX recurrent model is the best network structure in thermal analysis of dry type transformer windings.

905

Dry type transformer , artificial neural network , dynamic modeling , recurrent

Journal Of The Faculty Of Engineering an‎d Architecture Of Gazi University

913

Journal Of The Faculty Of Engineering an‎d Architecture Of Gazi University