Transformer hot spot temperature prediction using a hybrid algorithm of support vector regression and information granulation

摘要

A novel algorithm for transformer hot spot temperature prediction is proposed and presented in this paper. The algorithm is an integration of Support Vector Regression (SVR) and Information Granulation (IG), which is based on the principle of time series regression. The historical records consisting of measured hot spot temperature, top oil temperature, load current and ambient temperature of a transformer are used for verifying the proposed hybrid algorithm. The results show that the algorithm consistently outperforms a number of existing thermal modelling based methods (IEEE model, Swift's model and Susa's model) in estimating transformer's hot spot temperature.