Advanced VTL model for accurate prediction of composite THz metamaterial absorber frequency response

The design of terahertz (THz) metamaterial absorbers has evolved from simple structures to complex composites to meet the demands for multi-frequency absorption, broadband absorption, and polarization independence. However, the nonlinear interactions and expanded design spaces of composite structures pose significant challenges, making traditional design methods time-consuming and labor-intensive. To address these issues, this study proposes a data-driven framework that integrates a hybrid variational autoencoder-transformer-long short-term memory (VTL) model. The architecture is specifically designed to capture structural-spectral relationships, where the transformer models global dependencies among structural parameters, the long short-term memory (LSTM) network enhances the modeling of sequential spectral features in the terahertz frequency range, and the variational autoencoder (VAE) improves feature representation by learning implicit latent distributions. This integrated design enables effective characterization of the complex electromagnetic responses of composite metamaterials. The proposed approach achieves high prediction accuracy, with a mean squared error (MSE) of 0.0009, a coefficient of determination (R²) of 0.9725, and a mean absolute error (MAE) of 0.0161. It predicts a perfect absorption rate of 99.9% and optimally adjusts structural parameters to achieve targeted frequency responses. By addressing the limitations of traditional methods, this framework not only shortens the design cycle and reduces experimental costs but also offers a robust solution for the efficient design of high-performance THz metamaterial absorbers.