Accurate Antenna Design by Deep Auto-encoder Surrogate Model Assisted Particle Swarm Optimization

Abstract

Particle swarm optimization (PSO) is an evolutionary algorithm based on swarm cooperation and can deal with arbitrary optimization problems effectively. Therefore, PSO has been widely used in antenna design. Traditional antenna design approach (directly using PSO with an electromagnetic (EM) simulator) requires a massive amount of EM simulations during each optimization iteration, which is very time consuming. To address this situation, a novel approach using deep convolutional auto-encoder (DC-AE) assisted particle swarm optimization is proposed to design antenna. Using the antenna data collected by EM solver to train d-ae, the surrogate model is established. The surrogate model can predict the response of antenna unknown physical parameters quickly and accurately, which accelerates the iterative process of PSO. An antenna design example is used to demonstrate the proposed technique.