Machine Learning-Assisted Optimization of Direction-Finding Antenna Arrays

Antenna design is usually a complex optimization problem. It usually includes two stages. The first stage is to formulate it as an optimization problem. And then the second stage is to design an algorithm to solve the optimization problem. This paper aims to design unequally spaced antenna arrays for direction-finding, which requires cooperation among direction-finding users, antenna designers and optimizers. The paper combines the direction finding array with the optimization design and analyzes how to construct a reasonable optimization problem from the practical optimization design problem construction stage which has been neglected. A multi-objective problem is constructed by taking the direction-finding performance as the optimization objective. The mutual coupling effect causes the design of array antennas to require electromagnetic simulation to ensure the reliability of the results, which is a very time-consuming and expensive problem. Gaussian regression model is introduced into the multi-objective optimization algorithm to construct inexpensive surrogate optimization problems that reduce the number of accurate simulation evaluations. The experimental results show that the designed array meets all the requirements with good robustness while the number of simulations required is less than that of previous methods.