Multi-Objective Design of Compact Microwave Components with Data-Driven Surrogates and Pareto Front Decomposition

The paper discusses low-cost multi-objective optimization of compact microwave components using variable-fidelity EM simulation models and data-driven surrogates. Our approach builds upon a recently reported method where the initial approximation of the Pareto set is obtained by optimizing the kriging surrogate constructed from sampled data of the coarse-discretization EM model of the structure at hand, with selected designs further refined to obtain the high-fidelity Pareto set. The drawback of the method is a large number of training data samples required to set up the surrogate. Here, considerable savings concerning the training data set size are achieved by Pareto front decomposition based on auxiliary points identified along the front and setting up the kriging models in the corresponding sub-domains. The key factor is that the total volume of the sub-domains is considerably smaller than the volume of the original domain. Our considerations are illustrated using a compact rat-race coupler with design optimization cost savings of 29- and 30-percent for two and three sub-domains, respectively.