This study presents a data-driven symbolic regression framework for modeling power electronic converters, targeting applications in digital-twin environments. The proposed method automatically uncovers both the structural form and parameter values of governing equations directly from measured or simulated data, producing concise and interpretable discrete-time models capable of reproducing nonlinear switching behavior. A non-ideal boost converter is used as a case example, where conduction and switching losses are explicitly included to ensure a realistic dynamic representation. In contrast to conventional black-box machine learning approaches, the proposed strategy offers high computational efficiency and maintains strong predictive capability even when trained on relatively small datasets. The resulting models exhibit close agreement with physics-derived baselines, demonstrating the approach’s accuracy, scalability, and suitability for power electronics applications.
