Multibranch Machine Learning-Assisted Optimization and Its Application to Antenna Design

Abstract

Many full-wave electromagnetic (EM) simulations are needed to design an antenna meeting certain requirements, which involves a considerable computational burden. A multibranch machine learning-assisted optimization (MB-MLAO) method is proposed to dramatically reduce the computational complexity involved in this task. This method is then applied to antenna design and worst-case performance (WCP) searching under a practical manufacturing tolerance. In the conventional Gaussian process regression (GPR)-based MLAO method, a lower confidence bound (LCB) prescreening strategy with an empirical LCB constant is used to weigh the predicted value and predicted uncertainty. Using a variable-fidelity machine learning method, an adaptive LCB variable, and a retraining and repredicting method, the proposed MB-MLAO method can strike a delicate balance between exploitation and exploration in searching. Moreover, variable-fidelity data from full-wave EM simulations are used in the deep GPR machine learning method to further reduce the computational burden. Finally, two test functions and four types of antennas are selected as examples to illustrate the superiority of the proposed MB-MLAO method.

Application of Machine Learning-Assisted Global Optimization for Improvement in Design and Performance of Open Resonant Cavity Antenna

“…They are computationally cheap and can approximate the antenna behavior replacing the expensive feature of optimization in the commercial full-wave simulators. Nowadays, several ML-assisted antenna optimization methods are available [39], [56]− [60]. However, many of them are not very generic, are reliant on many ad-hoc processes, or have limitations on the number of design variables, or the range for surrogate modeling and search.…”

Automated Design of Antennas Using AI Techniques: A Review of Contemporary Methods and Applications

“…The general idea behind multifidelity optimization of antennas is to filter out non-promising design solutions using low-fidelity models that are inexpensive to simulate but less accurate, and to search around "promising" solutions discovered by the low-fidelity model using more accurate and expensive high-fidelity models. The models could be surrogate and/or EM models [26], [27], [28]. Multifidelity optimization methods have been applied to several antenna design problems.…”

“…Multifidelity optimization methods have been applied to several antenna design problems. For example in [28], an ultrawide band monopole antenna, a dual-band monopole antenna, a triband patch antenna, and a series-fed microstrip array antenna have been designed using this approach. The method in [28] improves the conventional Gaussian process regression (GPR)-based ML-assisted optimization of antennas via a multi-branch approach involving the use of multiple fidelity models to generate multifidelity GPR models and multiple constants or thresholds for the lower LCB prescreening.…”

An efficient parameter-retrieval-based surrogate-assisted optimization of on-platform honeycomb absorbing structures

“…Wang et al [32] presented an efficient GP-based method to analyze the EM scattering of 3D objects with varying shapes. Chen et al [33] proposed a multi-fidelity GP regression ML method to design the antenna. Moreover, multi-fidelity models have been employed in optimization studies [34].…”