Model-Agreement-Aware Multi-Objective Optimization for High-Frequency Transformers in EV Onboard Chargers

Developments in electric vehicle (EV) technology are pushing on-board chargers (OBCs) toward higher power density and efficiency, making high-frequency transformer loss prediction a critical design bottleneck. However, the accuracy of commonly used analytical winding-loss models varies strongly with frequency, conductor type (Litz/solid), window fill factor, and winding layout (e.g., interleaved), which can render single-model-based optimization unreliable. In this study, six analytical copper-loss models from the literature were independently reimplemented in a unified Python 3.11.5 workflow with a standardized interface to enable fair comparison under identical geometry and operating conditions. The models were benchmarked against 2D finite-element simulations on test scenarios with increasing physical complexity, including high fill-factor Litz windings and interleaved arrangements. The results confirm a regime-dependent behavior: no single model consistently outperforms others across the full design space, and model dispersion increases in geometrically stressed and higher-frequency regions. To manage this uncertainty, variance maps were generated and model disagreement was quantified using the coefficient of variation (CV). Finally, a reliability-oriented multi-objective optimization framework based on NSGA-II was developed, where a SmartTransformerRouter selects a reference loss estimate per candidate and CV is incorporated via constraints/penalties, with optional FEM triggering in high-uncertainty regions.