Physical, semi-physical and computational fluid dynamics thermal models of power transformers using artificial neural networks - A review

This paper emphasizes the importance of accurate measurement and monitoring of transformer temperatures, especially the hot spot temperature, because it directly affects the useful life span and operational reliability of the transformer. The thermal models are categories into physical, semi-physical and computational fluid dynamics (CFD) models. Advantages and limitations of each model are proposed. The role of neural networks in improving the accuracy of thermal predictions in semi-physical models is examined. The accuracy of different neural network models, including static models (multilayer perceptron, fuzzy neural network, radial basis function) and recurrent models (network neural element, recurrent fuzzy neural network) deals in predicting the thermal behavior of transformers.