Machine Learning-Based Estimation of Frequency-Dependent Transformer Parameters for Fast-Front Transient Studies

Abstract

Accurate frequency-dependent inductances and resistances are essential for high-frequency transformer models. Traditional analytical approaches, such as cases where eddy-current losses are neglected, or resistances and inductances are computed independently, and numerical techniques such as finite element methods (FEM) are either computationally intensive or rely on simplifications that reduce accuracy. This letter proposes a novel machine learning (ML)-based approach to efficiently estimate these parameters by learning from detailed analytical results. Using a localized feature selection strategy with conductors near the nearest neighbors $k$, the model considers complex electromagnetic interactions while achieving a significant reduction in computation time. This allows for generalization across different winding designs, reducing the dependence on traditional simplifications. Furthermore, the trained ML model achieves high accuracy, with predictions within an error margin of 5% for a wide frequency range. Comparison with measurements confirms the validity and effectiveness of the proposed approach, making it a promising solution for electromagnetic transient simulations.