Soft Computing Techniques to Model the Top-oil Temperature of Power Transformers

Abstract

This paper presents an investigation and a comparative study of four different approaches namely ANSI/IEEE standard methods, Adaptive Neuro-Fuzzy Inference System (ANFIS), Multilayer Feedforward Neural Network (MFNN) and Elman Recurrent Neural Network (ERNN) to modeling and prediction of the top-oil temperature for the 8 MVA Oil Air (OA)-cooled and 27 MVA Forced Air (FA)-cooled class of power transformers. A comparison of the proposed techniques is presented for predicting top-oil temperature based on the historical data measured over a 35 day period for the first transformer and 4.5 days for the second transformer with either a half or a quarter hour sampling time. Comparison results indicate that hybrid neuro-fuzzy network is the best candidate for the analysis and predicting of power transformer top-oil temperature. The ANFIS demonstrated the paramount performance in temperature prediction in terms of Root Mean Square Error (RMSE) and peaks of error.