Coarse models for efficient space mapping optimisation of microwave structures

Abstract

It follows from both theoretical results and practical observations that the coarse model is one of the most critical components of the space mapping optimisation process, affecting both the algorithm's ability of finding a high-quality design, and its computational complexity. A good coarse model should be a good representation of the fine model and, at the same time, it should be computationally cheap. The first property not only ensures the quality of the final design but also good convergence properties of the algorithm, so it also affects the computational complexity of the optimisation process through reducing the number of fine model evaluations required to find the solution. The second property ensures that the overhead related to parameter extraction and surrogate optimisation is small or even negligible. This study discusses techniques for creating computationally cheap and reliable coarse models. The approaches the authors present include interpolated models, multi-coarse-model techniques and the use of built-in capabilities of the coarse model simulator. The authors provide examples involving microwave design optimisation problems.