Composite Merit-Oriented Antenna Design Using Gray-Box Machine Learning-Assisted Optimization

Existing data-driven optimization methods struggle to address composite-merit antenna design problems efficiently. In this paper, an efficient grey box machine-learning-assisted optimization (GB-MLAO) method for composite-merit antenna design is proposed. The method utilizes a grey-box probabilistic modeling method to introduce multiple high-quality composite acquisition functions tailored for composite merit-oriented design, subsequently employing a batched potential design search strategy incorporating k-medoids clustering. Compared to conventional MLAO methods, GB-MLAO facilitates accurate modeling of diverse antenna merits, enhances exploration capability within the design space, and offers potential for parallel validation. The efficacy of the proposed method is assessed using benchmark test functions. Furthermore, GB-MLAO’s performance is demonstrated through three practical antenna design challenges: a tri-band slot antenna, a filtenna, and a heatsink-integrated base station antenna, targeting merits ranging from electromagnetic to thermal performance. Comparisons indicate that GB-MLAO possesses significant advantages over conventional methods.