A Trust-Region Parallel Bayesian Optimization Method for Simulation-Driven Antenna Design

Abstract

This article presents a trust-region parallel Bayesian optimization (TRPBO) method for simulation-driven antenna design problems. In this method, a multipoints acquisition function and trust region (TR)-based design space were developed to accelerate the solution efficiency of antenna design optimizations. For each optimization cycle in TRPBO, the acquisition function is applied to choose multiple updating points in a dynamic sampling space managed by a gradient-free TR, and then the responses at the updating points are calculated using parallel electromagnetic simulations. The proposed TRPBO was illustrated and tested on several benchmark problems, and then the TRPBO was applied to perform the design optimizations of a circularly polarized antenna, two wideband antennas, and an array antenna. Compared with the state-of-the-art methods, the proposed TRPBO achieves better optimization results with significantly less calculation time. The measured results of three fabricated antenna prototypes further show great potential of the TRPBO for solving simulation-driven antenna design problems.