H 2 -Optimal Parameteric Reduced-Order Model for Fast and Accurate Thermal Analysis of Power Modules Under Varying Cooling Boundaries

Precise temperature monitoring is crucial for ensuring reliability in high-power energy conversion systems, particularly those utilizing power modules. Lumped parameter thermal network (LPTN) model is limited by oversimplification, complex identification and boundary correlation, while finite element method (FEM) is too complicated for practical application. To address these challenges, this paper introduces a parametric model order reduction (PMOR) framework based on matrix projection interpolation (MPI). After extracting the state space representation from a high-fidelity FE model, a Krylov subspace projection with an iterative eigenvalue scheme is employed to identify optimal frequency expansion points, achieving an H2-optimal reduced model. A subspace reference basis projection further aligns reduced-order state variables under varied convective heat transfer boundaries, allowing smooth matrix interpolation without repeated model recalibration. As implemented on a commercial half-bridge power module, the proposed PMOR reduces the system order from 10340 to 10 with experimental temperature prediction errors within 1.99%, while requiring only 0.005% of the original computation time. Comparative experiments under different convective heat transfer boundaries and power losses validate the method's effectiveness and highlight its utility for thermal management in power modules.