Prediction and Optimization of Magnetic Core Loss Driven by Multifactor Coupling

ABSTRACT Accurate prediction of magnetic core loss is paramount for advancing power electronics but is severely hindered by the nonlinear coupling of multiple factors including frequency, flux density, excitation waveform, temperature, and material properties. To address this, we propose a comprehensive data‐driven framework that integrates model refinement, coupling analysis, and multiobjective optimization. First, a temperature‐corrected Steinmetz equation is developed, extending its validity and reducing the mean relative prediction error by more than half. Second, the individual and synergistic effects of these key variables are quantified via regression and the Artificial Hummingbird Algorithm (AHA), pinpointing the optimal combination for global loss minimization. Third, an ultra‐accurate Gradient Boosting Decision Tree (GBDT) model is established as a fast surrogate, leveraging key features like peak flux density for subsequent optimization. Finally, a constrained multiobjective optimization is performed, balancing core loss minimization against magnetic energy transfer maximization. The derived optimal design achieves a 30% core loss reduction compared to conventional benchmarks without compromising energy throughput. This work provides a systematic, data‐driven pathway for the design of high‐performance magnetic components, offering significant performance gains over traditional approaches.