A Machine Learning-Assisted Mixed-Integer Algorithm for Antenna Design Optimization - RMGS

Description

Designing Antennas can be very complex and computationally expensive, generally requiring evaluations of numerous design configurations to obtain the optimal solution. Traditionally, only continuous variables have been considered for artificial intelligence (AI)-based antenna optimization algorithms. For some antenna configurations, however, discrete variables combined with continuous variables are needed to be considered. Thus far, however, only sparse attention has been paid to mixed integer optimization.In this project, a novel algorithm based on machine learning (ML) and swarm intelligence is proposed for solving the hybrid-variable antenna problem efficiently. In detail, a variant of the radial basis function (RBF) network will be investigated for predicting the antenna responses. Then, the RBF networks will be tightly integrated with a mixed-variable version of particle swarm optimization (PSO). Overall, the proposed algorithm will tightly integrate the PSO with the RBF model in three parts: the RBF-assisted global search, the RBF-guided offspring generation strategy, and the prescreening strategy. This algorithm will lead to more efficient and effective antenna design techniques, enabling designs of higher-performance antennas at lower costs. It should be mentioned that our proposed approach can be applied to a wide range of industrial applications, such as electronic design automation (EDA).