Inverse Design of Dual-Frequency Reflective Metasurfaces Based on Deep Learning and Multi-Objective Optimization Algorithms

This paper presents a novel inverse design method integrating deep learning models with multi-objective optimization algorithms to achieve rapid inverse design of metasurface elements. The method enhances the model's forward prediction performance by adopting Kolmogorov-Arnold Networks (KAN) instead of the traditional Multi-Layer Perceptron (MLP). It also employs the Non-dominated Sorting Genetic Algorithm II (NSGA-II) to address the uncertainty issue in non-target frequency bands and the problem of multiple design objectives not being able to simultaneously reach optimality in dual-frequency metasurface element design. Based on this method, A dual-frequency reflectarray antenna we designed achieves symmetric dual-beam and single-beam deflection at the frequencies of 8.0 GHz and 16.5 GHz, respectively. The results demonstrate that this method not only efficiently realizes the on-demand inverse design of dualfrequency metasurface elements but also provides an innovative solution for the design of complex metasurface structures.