Design of Frequency-Selective Rasorbers Based on a Hybrid Neural Network Framework

This letter presents an intelligent design method for frequency selective rasorbers (FSRs) based on a hybrid neural network framework, significantly reducing learning complexity. First, a hybrid modeling approach is introduced, integrating topological modeling for the absorber layer and parametric modeling for the frequency selective surface (FSS) layer, thereby enhancing efficiency and flexibility. Next, a neural network framework is constructed using a variational autoencoder and a hybrid predictor to establish relationships between electromagnetic properties and FSR modeling parameters. The predictor consists of two sub-predictors, tailored for the absorber and FSS layers, respectively, and a combiner that integrates their outputs. Finally, the trained predictor is coupled with genetic algorithm to enable intelligent FSR design. Numerical examples demonstrate the practicality and accuracy of the proposed method.