Estimation of temperature development in secondary substation transformers

Abstract

The temperature in a transformer element has a great impact on the aging process fora transformer. After a pilot project that aimed to make their power grid smarter, Agder Energi had temperature-data from sensors placed on some of their transformers. The purpose of this thesis was to establish a method for estimating transformer temperature based on available information such as load, capacity, and external weather conditions. To succeed, we had to deal with limited data, that is, few relevant samples combined with high levels of noise. To address the challenges of the estimation problem and the limitations of the data, we proceeded as follows. We used XGBoost on the dataset, which was resampled to put importance on the suitable samples. Feature engineering was done to improve the model. Optimization of the XGBoost parameters improved performance further. Training on 85%and testing on 15% of the data, we obtained an RMSE of3.0. Additional results is presented, and the reliability of the results is discussed with consideration to the limitations. Agder Energi will take the project further, where it will be put into production to provide valuable insights.