Comparison of Recurrent Neural Networks on Dry-type Transformer Thermal Models under Various Conditions

The rising of Electric Vehicles (EV) can enormously affect transformers, which are major
components of power systems. Non-linear loads cause significant number of harmonic components and additional losses in transformers. The rising hot-spot transformer temperature also increases along with the losses. Thermal condition monitoring applied to the transformers can prevent the transformer overheating and failure in the power systems. In this work, thermal models for hot-spot temperature prediction are implemented based on the Recurrent Neural Network (RNN) with an experimental time-series dataset. The RNN has ability to learn thermal characteristics of dry-type transformers.