Multi-Output Variable-Fidelity Bayesian Optimization of a Common Mode Choke
Rodrigo Silva Rezende, Mirsad Hadžiefendić, Jan C. Hansen, Rolf Schuhmann · 2021 IEEE International Joint EMC/SI/PI and EMC Europe Symposium · 2021
In this paper, we propose a modification of the well-known kriging modeling technique and its multi-fidelity variant in order to generate a surrogate model of a vector-valued output of a common mode choke. Combined with Bayesian optimization, we solve a multi-objective optimization problem to find the tradeoff between inductance, total volume, and resonance frequency. With the proposed method, we calculate a Pareto front of the objective functions more than 200 times faster compared to the solution without the surrogate.