Accelerating optimization of terahertz metasurface design using principal component analysis in conjunction with deep learning networks

Abstract

Metamaterials are a class of artificial materials that have exceptional physical properties that do not exist in nature. They are widely used in various fields, such as electromagnetics, optics, and acoustics. However, designing metamaterials can be a challenging and time-consuming task. Traditional methods rely on simulations and trial-and-error, which are inefficient and often require significant computational resources. Recently, deep learning has emerged as a promising tool to design metamaterials. Deep learning involves training neural networks to learn complex patterns and relationships in data, which can be used to predict the behavior of metamaterials under different conditions. This paper proposes a neural network that maps geometric parameters to frequency domain responses for optimized design. The network utilizes PCA (Principal Component Analysis) to reduce the training time by approximately 5%, and this combination method is far superior to similar algorithms in terms of prediction accuracy and generalization ability. Experimental results demonstrate that the designed network model can be used for optimized design, achieving a remarkably low RMSE (Root Mean Square Error) of 0.0408 and a prediction accuracy of 97.64% in the reverse network, outperforming similar articles. The proposed network model improves the design efficiency of metamaterials, providing a more efficient and effective approach for designing these metamaterials.