A Single-Layer Frequency Selective Surface Design Method Based on Multi-Layer Perceptron

In this paper, a single-layer frequency selective surface (FSS) design method based on multi-layer perceptron (MLP) is proposed. The feature parameters of FSS are randomly sampled at first and the corresponding electromagnetic (EM) responses are collected by full-wave simulations. Then the MLP is trained to predict the EM response with new parameters instead of the full-wave simulation. To verify the proposed design method, a single-layer frequency selective surface is designed, consisting of a multi-loop structure where 2000 data sets are used to train the MLP model. The realized mean absolute error is 0.0121. The average time consumption is reduced from 27.72 seconds to 0.00178 seconds compared with the full-wave simulation. This method is a good candidate to build the EM agent model.