Bayesian Optimization for Antenna Design via Multi-Point Active Learning

Abstract

In recent years, Bayesian optimization (BO) has shown promising results in electromagnetic designs. However, the existing methods only chose one updating point per optimization cycle, which is time-wasting when parallel computation can be used. To address this problem, a new method, called parallel Bayesian optimization (PBO) is presented in this letter. In this method, a multi-point acquisition function was developed to accelerate the solution efficiency of antenna design optimization. For each optimization cycle in PBO, the acquisition function is applied to choose multiple updating points, and then the responses at the updating points are calculated in parallel. The proposed PBO was applied to design an ultra-wideband microstrip antenna. Compared with the state-of-the-art methods, the PBO achieves better optimization results with less calculation time. The measured results of the antenna prototypes further show great potential for solving antenna design optimization problems.