A Novel Method for Frequency Selective Surface Design Using Deep Learning with Improved Particle Swarm Algorithm

Abstract

This paper presents a design method for frequency selective surface (FSS) based on the deep neural network and improved particle swarm algorithm (IPSO). In the proposed method, the forward prediction network (FPN) based on the fully connected network is established to fast predict the transmission coefficient of FSS. Combined with the FPN, the IPSO is used to optimize the structural parameters of FSS. Compared with the traditional iterative optimization method based on full-wave simulation, this method greatly improves the optimization efficiency of FSS. For example, a band-stop FSS is optimized with the proposed method in 210.6s, and the optimization efficiency increases by more than 99%. Simulation results show that the transmission coefficient errors of key frequency points between optimization results and objectives are less than 1 dB. And the deviation of the center frequency and the bandwidth of the target frequency bands is less than 0.81% and 4.1%, respectively.