Design of Beamforming Metasurface Based on Fully Connected Neural Networks

In order to solve the problem that the period of designing metasurface with specific functions is too long by using electromagnetic simulation software, this paper proposes a metasurface design method based on artificial neural network, which can accurately predict the far-field pattern. In the training process, the root-mean-square error decreases with the increase of the number of iterations, the minimum root-mean-square error is 0.01, and the training time is 21 s. After the training is completed, the model can quickly predict the far-field pattern according to the coding matrix. The single prediction time of the neural network is 0.0065 s, while the single simulation time of the full-wave simulation software is 4 min, which is far more than that of the neural network. The accuracy of the predicted data is verified by comparing the predicted results with the results of the full-wave simulation. By combining the neural network with particle swarm optimization algorithm, the metasurface coding matrix corresponding to the target pattern can be designed. Compared with the traditional metasurface design method, this design method greatly reduces the design time and solves the problems of time consuming and cost in the traditional design method.