Utilizing Kriging Surrogate Models for Multi-Objective Robust Optimization of Electromagnetic Devices

Abstract

This paper presents a multi-objective robust optimization strategy assisted by the surrogate model. In order to guarantee the accurate response prediction, the performances of three different Kriging surrogate models, ordinary Kriging, first-order universal Kriging (UK), and second-order UK, are investigated through analytical benchmark functions. Once the accurate model is constructed, the performance analysis can be efficiently approximated during optimization process. Furthermore, the robustness against uncertainty is evaluated by the worst-case scenario through applying optimization technique to the approximated model in the uncertainty set. The proposed algorithm is validated through one electromagnetic application, a robust version of the TEAM 22.