Machine-Learning-Assisted Optimization for Antenna Geometry Design

A machine-learning-assisted optimization (MLAO) method for antenna geometry design (AGD) (MLAO-AGD) is proposed. By combining machine learning (ML) methods, including a convolutional neural network (CNN) and Gaussian process regression (GPR), MLAO-AGD achieves great efficient improvement compared with conventional evolutionary-algorithm-assisted AGD methods. The ML methods are introduced to build surrogate models between the antenna geometry and the antenna performance and then provide predictions of potential designs during optimization. The ML-based surrogate model is iteratively updated by verified optimization results using full-wave simulations. Three antenna design examples, including multiband and broadband antenna element design tasks and a mutual coupling reduction design task, are presented to show the advantages of the proposed MLAO-AGD, which include the convergence speed and antenna performance.