航天器环境工程2025，Vol.42Issue(5)：468-476,9.DOI:10.12126/see.2025068
基于深度学习的电磁超材料设计方法及性能验证
Deep learning-based design and performance validation of electromagnetic metamaterials
雷晗 1胡奇 1叶田园 1何晶 1许亚娟 1白长行 1邵丽娟1
作者信息
- 1. 北京卫星环境工程研究所,北京 100094
- 折叠
摘要
Abstract
To address the challenge of balancing performance and efficiency in the design of electromagnetic metamaterials,a joint optimization framework integrating a Residual Fully-Connected Network(RFCN)with an Improved Genetic Algorithm(IGA)is proposed.First,an RFCN model was constructed to efficiently predict the reflectance curves of a stepped-cone lattice structure in the 2-40 GHz frequency band(test set RMSE=0.38).Subsequently,an IGA incorporating catastrophic mutation,large-scale mutation,and precise screening mechanisms was introduced for global parameter optimization.Simulation and experimental results showed that the number of convergence generations was reduced from 36 to 14,indicating a significant improvement in optimization efficiency.The optimized structure achieved a reflectance below-10 dB in the 3.4-36.3 GHz and 39-40 GHz bands,with effective absorption bandwidths of 91.2%(simulation)and 89.3%(measurement),respectively.Significant absorption peaks were observed at 3.79,15.5,and 35.3 GHz,with a minimum reflectance of-41.77 dB.The proposed method overcomes the limitations of traditional physics-based design and provides an efficient and accurate approach for electromagnetic metamaterial design,demonstrating strong potential for applications in spacecraft stealth and electromagnetic compatibility design.关键词
电磁超材料/深度学习/遗传算法/吸波性能/结构优化Key words
electromagnetic metamaterials/deep learning/genetic algorithm/absorption performance/structural optimization分类
通用工业技术引用本文复制引用
雷晗,胡奇,叶田园,何晶,许亚娟,白长行,邵丽娟..基于深度学习的电磁超材料设计方法及性能验证[J].航天器环境工程,2025,42(5):468-476,9.
