Model management for cost-efficient surrogate-based optimisation of antennas using variable-fidelity electromagnetic simulations
Slawomir Marcin Koziel, Stanislav Ogurtsov · IET Microwaves Antennas & Propagation · 2012
Electromagnetic (EM) simulation has become an important tool in the design of contemporary antenna structures. However, accurate simulations of realistic antenna models are expensive and therefore design automation by employing EM solver within an optimisation loop may be prohibitive because of its high computational cost. Efficient EM-driven antenna design can be performed using surrogate-based optimisation (SBO). A generic approach to construct surrogate models of antennas involves the use of coarse-discretisation EM simulations (low-fidelity models). A proper selection of the surrogate model fidelity is a key factor that influences both the performance of the design optimisation process and its computational cost. Despite its importance, this issue has not yet been investigated in the literature. Here, the authors focus on a problem of proper surrogate model management. More specifically, the authors carry out a numerical study that aims at finding a trade-off between the design cost and reliability of the SBO algorithms. Our considerations are illustrated using several antenna design cases. Furthermore, the authors demonstrate that the use of multiple models of different fidelity may be beneficial to reduce the design cost while maintaining the robustness of the optimisation process. Recommendations regarding the selection of the surrogate model coarseness are also given.