Surrogate-assisted inverse design of random metasurfaces using a complementary dual-population genetic algorithm
Journal of the Optical Society of America B · 01.12.2026 · 43 (12) · 2941
DOI: 10.1364/josab.610011 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of the Optical Society of America B, online erschienen am 03.11.2026. Verfasst von Man Zhang und 2 weiteren Autor:innen. Die Arbeit nennt 41 Literaturangaben. Gefördert von: Science and Technology Planned Project of Guizhou Province, Nature Science Foundation of Guizhou Minzu University, Doctoral Research Launch Grants of Guizhou Minzu University.
Abstract (Originalsprache)
Metasurface inverse design enables flexible electromagnetic-wave manipulation, yet fully random-pattern metasurfaces with ultrahigh design freedom remain challenging owing to their strongly nonlinear geometry–response mapping: simulation-driven methods bear prohibitive cost, gradient-based optimizers converge to local optima, heuristic algorithms suffer premature convergence, and deep generative models degrade under the intrinsic one-to-many mapping. This work proposes a surrogate-assisted inverse-design framework centered on a complementary dual-population genetic algorithm (CDP-GA), in which a learned forward model replaces full-wave simulation for fitness evaluation. The adopted VAE-CNN-Bi-LSTM surrogate compresses the 2002-dimensional complex S-parameter spectrum into a 32-dimensional latent space; trained on 108,000 samples of the RDN subset (2 256 configurations), it reaches mean squared errors of 0.0093 (real) and 0.0072 (imaginary) on a 3000-sample test set, the lowest among the compared architectures. For the benchmark random-metasurface inverse-design tasks considered, through an “Elite–Potential” dual-population mechanism, the CDP-GA mitigates premature convergence and surpasses standard GA, pre-optimized GA, DE, and BPSO in solution quality and convergence stability on those tasks.
Förderung
- Science and Technology Planned Project of Guizhou Province
- Nature Science Foundation of Guizhou Minzu University
- Doctoral Research Launch Grants of Guizhou Minzu University
Zitieren
APA
Zhang, M., Zhu, W., & Long, F. (2026). Surrogate-assisted inverse design of random metasurfaces using a complementary dual-population genetic algorithm. Journal of the Optical Society of America B, 43(12), 2941. https://doi.org/10.1364/josab.610011
Vancouver
Zhang M, Zhu W, Long F. Surrogate-assisted inverse design of random metasurfaces using a complementary dual-population genetic algorithm. Journal of the Optical Society of America B. 2026;43(12):2941. doi:10.1364/josab.610011
BibTeX
@article{Zhang2026,
author = {Zhang, Man and Zhu, Weili and Long, Fei},
title = {Surrogate-assisted inverse design of random metasurfaces using a complementary dual-population genetic algorithm},
journal = {Journal of the Optical Society of America B},
year = {2026},
volume = {43},
number = {12},
pages = {2941},
doi = {10.1364/josab.610011}
}
EUVIMED