Inverse Model of Electromagnetic Metasurfaces via Physical Mechanism and Multilevel Collaborative Optimization

This letter presents an inverse model of electromagnetic metasurfaces (IM-EM) that integrates physical mechanism with a multilevel collaborative optimization framework for accurate and physically consistent metasurface inversion. Unlike conventional regression or purely iteration methods, IM-EM embeds physical constraints by transmission-line theory into a hybrid optimization strategy that combines genetic algorithm (GA) with sequential quadratic programming (SQP). The tailored optimization function balances amplitude, spectral trend, and physical regularization, ensuring stable convergence within a physically feasible solution space. The method achieves convergence within 20 iterations, yielding normalized root mean square error (NRMSE) below 0.05 and correlation coefficients above 0.98 across broadband 1–12 GHz. Robustness in additive Gaussian noise and different structures is confirmed, which illustrates the strong generalization for complex EM designs.