A survey of machine learning and evolutionary computation for antenna modeling and optimization: Methods and challenges

Antenna design is a kind of electromagnetic (EM) engineering problem and normally formulated as complex nonlinear optimization problem. Evolutionary computation (EC) was combined early to antenna design due to its powerful nonlinear optimization capability. Modern antenna design depends on EM simulation software to solve Maxwell equations, which is time-consuming and makes it nontrivial for application of EC in antenna design. Machine learning (ML) is widely used to accelerate antenna design by building surrogate model of EM simulation. However, existing surveys focus on one of these two artificial intelligence (AI) methods (EC and ML) in antenna applications, and have overlooked differences between two cases of surrogate model for antenna EM simulation (response modeling and specification modeling). This review paper aims to summarize the applications of both EC and ML in antenna design over the past decades and highlight advantages and disadvantages of two kinds of EM simulation surrogate models. The survey begins with a short overview of ML and EC basics. Then various applications are discussed in three parts, including antenna optimization with EC, ML-assisted antenna optimization with response modeling and ML-assisted antenna optimization with specification modeling. Finally, challenges and potential future directions for applying ML and EC in antenna design are discussed, as well as emerging trends. This survey provides a comprehensive introduction to ML and EC in antenna design and contributes to the investigation of AI-empowered antenna design.