Hot spot temperature estimation in mineral oil immersed power transformers using support vectors regression

Abstract

This article presents an innovative methodological development for the hot spot temperature estimation in mineral oil immersed power transformers by using the support vector regression (SVR). The SVR algorithm is based on the statistical learning theory and is part of the machine learning tools. It was used through a six stage implementation where an SVR model capable of estimating the variable under study is obtained. The method was applied to a real 30 MVA transformer with ONAN/ONAF cooling at 70/100 % of load using a database for a 10 year period. The developed SVR model was validated by comparison to the results obtained with the Dejan Susa model using statistical performance metrics. In conclusion, the results obtained indicate that the implemented SVR model allows estimating the hot spot temperature with high accuracy.