A Machine Learning Approach to Estimate Stray Losses in Power Transformers

Power transformers experience different types of energy losses that influence their overall efficiency. 2D simulations are classically applied in transformers industry to predict losses in transformer windings, and the use of 3D simulations to calculate losses in structural components has increased in recent years. Losses in certain structural components, such as the core, pose challenges for efficient and quick simulation during the design stage. In this study, specific formulas incorporating measurements of total load losses during factory acceptance tests are employed. Starting from a dataset of 185 samples of transformers, several machine learning techniques were tested to estimate the stray losses of the structural components, defined as losses that occur due to leakage field of windings and field of high current carrying leads/bus-bar. The overarching goal of the thesis is to assess the performance of the employed machine learning models in predicting these stray losses, especially in comparison with more traditional alternatives. An enhanced prediction of the losses would allow the company to save resources during the transformer design stage.