A Multi-Fidelity Surrogate Optimization Method Based on Analytical Models

In this work, an analytical method is proposed to derive the coarse model for a multi-fidelity surrogate optimization. Specifically, utilizing a set of eigenfunction expansions, that characterizes the solution domain of a desired geometry, the proposed surrogate model is trained along with the high-fidelity full-wave simulations. To demonstrate and validate the proposed framework, a loop antenna and a loop array are studied. Notably, the proposed multi-fidelity model is applied to optimize a multi-variable design and achieve two desired objectives. Utilization of eigenfunction expansions accelerates the multi-fidelity surrogate optimization method 200 to 300 times compared to conventional multi-fidelity surrogate models. As shown from our results, the proposed algorithm efficiently optimizes a loop antenna and loop array, obtaining a desired Pareto set of non-dominated designs.