Chapter 5: Fundamentals of Surrogate-Based Modeling and Optimization

Design optimization of antennas and antenna arrays is challenging for several reasons. One of these is that the process involves high-fidelity EM simulation models necessary for reliable performance evaluation of the structure under design. Another reason is a large number of designable parameters: these might be dimensions of radiating elements and feeds, parameters of substrates as well as excitation amplitudes and/or phases. Other challenges might include typically graded composition of the array structures, presence of connectors and radomes. Furthermore, global optimization is often necessary because functional landscapes pertinent to antennas and antenna arrays tend to be multi-modal with multiple local optima. All these challenges make conventional numerical optimization techniques (see Chapters 3 and 4 for details) unsuitable for handling real-world antenna design problems. In particular, the computational cost of such algorithms is typically measured in hundreds (for local methods) or thousands and tens of thousands (for population-based metaheuristics) of objective function evaluations. A notable exception is gradient-based search with adjoint sensitivities (Ghassemi et al., 2013; Koziel and Ogurtsov, 2012c; Koziel et al., 2013b, 2014b; Koziel and Bekasiewicz, 2015a, 2016b) where the optimization process can be conducted in reasonable time even for relatively large number of designable parameters. Unfortunately, adjoints are not yet widespread in computational electromagnetics, especially in terms of availability of this technology through commercial simulation software packages (CST, 2016; HFSS, 2016)…