搜索

x

留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

基于迁移学习的共形超构表面散射场高效智能计算方法

刘凯越 李腾耀 郑娜娥 田志富 蔡通 王彦朝 曹超华

引用本文:
Citation:

基于迁移学习的共形超构表面散射场高效智能计算方法

刘凯越, 李腾耀, 郑娜娥, 田志富, 蔡通, 王彦朝, 曹超华

Efficient intelligent calculation method for conformal metasurface scattering field based on transfer learning

Liu Kai-Yue, Li Teng-Yao, Zheng Na-E, Tian Zhi-Fu, Cai Tong, Wang Yan-Zhao, Cao Chao-Hua
PDF
HTML
导出引用
  • 共形超构表面具有灵活的外形, 便于适应复杂的平台结构, 在移动载体散射调控方面具有显著的应用优势. 但共形超构表面散射场电磁仿真时间周期长、优化复杂, 难以支撑共形超构表面的敏捷设计. 本文提出了一种基于迁移学习的共形超构表面散射场高效智能计算方法. 首先, 根据天线理论和全波电磁仿真在物理机理上的同一性, 预先用大量理论数据训练源域模型, 构建相位分布与散射场的初始映射模型; 然后, 将少量全波仿真训练数据作为目标域样本, 依托预训练、参数冻结和模型微调实现初始模型的有效迁移, 完成超构表面相位分布到散射场映射模型的构建; 最后, 针对不同的共形几何结构, 在迁移学习模型基础上进行了二次迁移. 结果表明, 本文所提方法的散射场计算时间较全波仿真下降3个数量级, 在少样本条件下应用迁移学习后散射场计算精度平均提升19.8%, 训练数据量仅占未迁移学习模型的42.9%, 数据集收集时间缩短50.1%, 同时在超构表面共形结构发生改变时具有一定泛化性.
    Conformal metasurfaces with flexible structures can fit complicated platforms and have obvious advantages in moving platforms scattering manipulations. However, the far-field simulation of conformal metasurface is time-consuming and difficult to optimize, thereby making the its agile design difficult. Here, an efficient and intelligent scattering field calculation method is proposed based on transfer learning for conformal metasurfaces. Firstly, according to the consistency in physical mechanism between antenna theory and full wave simulation, an initial mapping model between phase distribution and far-field of metasurface is constructed and pre-trained based on a large quantity of theoretical data in source domain. Secondly, by pre-training, parameter freezing and model fine-tuning, the far-field prediction model for full wave simulation is transferred and achieved successfully, based on a small quantity of full wave simulation data in target domain. Finally, the transfer learning model for far-field prediction is transferred once again for conformal metasurfaces with different structures. Results indicate that the proposed method only consumes 0.1% of full wave simulation time for conformal metasurface far-field calculation. In fewer samples, the model with transfer learning can improve the average accuracy by 19.8%, and the training data account for only 42.9% for the models without transfer learning, which reduces the training data collection time by 50.1%. Moreover, our far-field calculation method demonstrates good transfer performance between conformal metasurfaces with different structures.
  • 图 1  共形超构表面迁移学习框架

    Fig. 1.  Transfer learning framework of conformal metasurface.

    图 2  共型超构表面剖面结构和单元电磁特性 (a) 茹科夫斯基翼型函数曲线; (b) 共型超构表面结构, 其中xy方向设置为unit cell边界, z方向为open边界; (c) 超构单元结构及其仿真设置; (d) 超构单元的同极化反射相位和幅度

    Fig. 2.  The structure of conformal metasurface and electromagnetic response of meta-atom: (a) Жуковский airfoil function; (b) the structure of conformal metasurface, where unit cell conditions are applied in x and y directions and open condition is applied in z direction; (c) prototype of the meta-atom and simulation setup ; (d) reflection phase and amplitude of co-polarization.

    图 3  源域训练数据生成和模型构建 (a) 基于天线理论计算得到大量相位分布和散射场数据; (b) 预训练模型${\mathcal{M}_0}$的网络架构

    Fig. 3.  Training data generation and model network construction of source domain: (a) Numerous phase distribution and scattering data can be generated based on antenna theory; (b) paradigm of pre-trained network ${\mathcal{M}_0}$.

    图 4  全波仿真数据获取设置 (a) 联合自动化仿真流程; (b) 共形超构表面全波仿真设置

    Fig. 4.  Configurations of full wave simulations: (a) The flowchart of associated automation simulation; (b) the full wave simulation setup of conformal metasurface.

    图 5  迁移学习策略

    Fig. 5.  Transfer learning strategy.

    图 6  模型结构分析 (a) 卷积块数量取1, 2, 3时模型的平均测试损失值; (b) 全连接层数量为1, 2, 3, 4, 5时模型的平均测试损失值; (c) 激活函数分别为ReLU, ReLU6, ELU, Leaky ReLU和Soft Plus模型平均测试损失值

    Fig. 6.  Network structure analysis. Model average train loss with (a) convolution blocks number of 1, 2, 3, (b) fully connected layer number of 1, 2, 3, 4, 5, and (c) activation function of ReLU, ReLU6, ELU, Leaky ReLU and Soft Plus.

    图 7  模型超参数寻优. 测试集的平均损失随 (a) 学习率, (b) 批大小, (c) 迭代轮次变化曲线

    Fig. 7.  Model hyperparameter optimization. The average loss of the test set varies with (a) learning rate, (b) batch size, and (c) iteration rounds.

    图 8  迁移学习冻结层数对测试集损失值的影响.

    Fig. 8.  The impact of transfer learning frozen layers numbers on loss value.

    图 9  模型${\mathcal{M}_0}$, ${\mathcal{M}_1}$和${\mathcal{M}_2}$的平均损失值和效果验证 模型${\mathcal{M}_0}$(a), ${\mathcal{M}_1}$(b), ${\mathcal{M}_2}$(c)的损失值随迭代轮次的变化曲线以及两种测试算例下的拟合效果

    Fig. 9.  Loss value and verification of ${\mathcal{M}_0}$, ${\mathcal{M}_1}$ and ${\mathcal{M}_2}$: Loss over the epoch and prediction results under two testing instances of ${\mathcal{M}_0}$(a), ${\mathcal{M}_1}$(b), ${\mathcal{M}_2}$(c).

    图 10  迁移学习模型与全波仿真时间对比 (a) 结构1共形超构表面在软件中进行全波仿真计算散射场, 与利用模型${\mathcal{M}_1}$计算散射场的时间对数值; (b) 结构2共形超构表面在软件中进行全波仿真计算散射场, 与利用模型${\mathcal{M}_2}$计算散射场的时间对数值

    Fig. 10.  Time comparison of transfer learning model and full-wave simulation: (a) Logarithmic value of full-wave simulation time in CST and model ${\mathcal{M}_1}$ calculation time for conformal metasurface with structure 1; (b) logarithmic value of full-wave simulation time in CST and model ${\mathcal{M}_2}$ calculation time for conformal metasurface with structure 2.

    图 11  迁移学习模型和未迁移学习模型性能对比分析 (a) 模型精度随训练数据量的变化关系; (b) 数据集收集时间柱状图

    Fig. 11.  Performances comparative analysis of models with and without transfer learning: (a) Model accuracy varies with the amount of training data; (b) bar chart of dataset collection time.

    图 12  结构2共形超构表面散射场计算模型平均损失函数及效果验证 (a) 基于结构2全波仿真数据直接训练得到的远场计算模型${\mathcal{M}'_2}$(未迁移学习); (b) 基于结构1理论数据散射场计算模型${\mathcal{M}_0}$迁移训练得到的结构2共形超构表面远场计算模型${\mathcal{M}''_2}$(经历一次迁移学习); (c) 基于结构1迁移学习模型${\mathcal{M}_1}$迁移训练得到的结构2共形超构表面远场计算模型${\mathcal{M}_2}$(经历两次迁移学习)

    Fig. 12.  Average loss value and performance verifications of structure 2 conformal metasurface far-field prediction models: (a) Far-field prediction model ${\mathcal{M}'_2}$(without transfer learning) for structure 2 conformal metasurface, trained based on structure 2 conformal metasurface full wave calculation data; (b) far-field prediction model ${\mathcal{M}''_2}$ (with once transfer learning) for structure 2 conformal metasurface, transferred from the structure 1 conformal metasurface theoretical far-field prediction model ${\mathcal{M}_0}$; (c) far-field prediction model ${\mathcal{M}_2}$ (with twice transfer learning) for structure 2 conformal metasurface, transferred from the structure 1 conformal metasurface full wave calculation far-field prediction model ${\mathcal{M}_1}$.

    图 13  不同轮廓方程结构迁移学习效果 (a) 轮廓方程为圆弧的共形超构表面散射场预测模型${\mathcal{M}_{\text{c}}}$的平均损失值和效果验证; (b) 轮廓方程为的二次曲线的共形超构表面散射场预测模型${\mathcal{M}_{\text{q}}}$的平均损失值和效果验证. 其中左图中的插图表示结构1轮廓和圆弧、二次曲线结构对比图

    Fig. 13.  The transfer learning performance of different contour equation structures: (a) Average loss value and performance verifications of far-field prediction model ${\mathcal{M}_{\text{c}}}$ for conformal metasurface with contour equation of a circular arc; (b) average loss value and performance verifications of far-field prediction model ${\mathcal{M}_{\text{q}}}$ for conformal metasurface with contour equation of a quadratic curve. The illustration in the left figure represents the comparison between the outline of structure 1 and the structure of arc and quadratic curves.

    表 1  模型结构的搜索空间

    Table 1.  Search space of model structure.

    序号控制变量候选集合
    1卷积块数量$ \{1, 2, 3\} $
    2全连接块数量(不含输出层)$ \{1, 2, 3, 4, 5\} $
    3激活函数类型{ReLU, ReLU6, ELU, LeakyReLU, SoftPlus}
    下载: 导出CSV

    表 2  模型超参数的搜索空间

    Table 2.  Search space of model hyperparameters.

    序号控制变量候选集合
    1学习率{0.0001, 0.0002, . .., 0.0099, 0.01}, 步长为0.0001
    2批量大小{32, 64, 128, 256, 512, 1024}
    3迭代轮次{5, 10, 15, 20, . .., 90, 95, 100}, 步长为5
    下载: 导出CSV

    表 3  模型迁移关系

    Table 3.  The transfer relationship between models.

    序号模型预训练模型迁移次数
    1${\mathcal{M}_0}$0
    2${\mathcal{M}_1}$${\mathcal{M}_0}$1
    3${\mathcal{M}'_2}$0
    4${\mathcal{M}''_2}$${\mathcal{M}_0}$1
    5${\mathcal{M}_2}$${\mathcal{M}_1}$2
    下载: 导出CSV
  • [1]

    Liu K Y, Wang G M, Cai T, Guo W L, Zhuang Y Q, Liu G 2018 Chin. Phys. B 27 084101Google Scholar

    [2]

    Liu K T, Wang G M, Cai T, Dai B J, Xiao Y, Li H P, Guo W L 2019 J. Phys. D: Appl. Phys. 52 255002Google Scholar

    [3]

    Bai H, Wang G M, Zou X J , Xie P, Shi Y P 2022 Chin. Phys. B 31 54102

    [4]

    Liu K Y, Wang G M, Cai T, Li H P, Li T Y 2021 IEEE Trans. Antennas Propag. 69 3349Google Scholar

    [5]

    Germain D, Seetharamdoo D, Burokur S N, de Lustrac A 2013 Appl. Phys. Lett. 103 124102Google Scholar

    [6]

    Wang Y C, Feng Q, Kong X D, Liu H X, Han J Q, Li L 2024 IEEE Antennas Wirel. Propag. Lett. 23 970Google Scholar

    [7]

    Fu B, Yu S X, Kou N, Ding Z, Zhang Z P 2022 Chin. Phys. B 31 40703Google Scholar

    [8]

    Gupta E, Bonner C, Lazarus N, Mirotznik M S, Nicholson K J 2023 IEEE Antennas Wirel. Propag. Lett. 22 2629

    [9]

    Fu C F, Han L F, Liu C, Lu X L, Sun Z J 2022 IEEE Trans. Antennas Propag. 70 2352

    [10]

    Wang Y J, Su J X, Li Z R, Guo Q X, Song J M 2020 IEEE Antennas Wirel. Propag. Lett. 19 631

    [11]

    Luo X Y, Chen K, Guo W L, Feng Y J 2019 International Conference on Microwave and Millimeter Wave Technology (ICMMT) Guangzhou China, May 19-22, 2019 pp 1

    [12]

    Xu H X, Hu G W, Wang Y Z, Wang C H, Wang M Z, Wang S J, Huang Y J, Genevet P, Huang W, Qiu C W 2021 Light Sci. Appl. 10 75Google Scholar

    [13]

    王朝辉, 许河秀, 逄智超, 王明照, 王少杰 2022 红外与毫米波学报 41 210Google Scholar

    Wang C H, Xu H X, Pang Z C, Wang M Z, Wang S J 2022 J. Infrared Millimeter Waves 41 210Google Scholar

    [14]

    Srilatha K., Madhav B. T. P., babu B. Anil, Das Sudipta, Rao M. C. 2022 Appl. Phys. A 128 1091Google Scholar

    [15]

    刘梦蛟, 李添悦, 戈钦, 王漱明, 王振林, 祝世宁 2022 光学学报 42 2126004

    Liu M J, Li T Y, Ge Q, Wang S M, Wang Z L, Zhu S N 2022 Acta Opt. Sin. 42 21260049

    [16]

    王朝辉, 许河秀, 逄智超, 王明照, 王少杰 2022 红外与毫米波学报 41 8Google Scholar

    Wang Z H, Xu H X, Pang Z C, Wang M Z, Wang S J 2022 J. Infrared Millimeter Waves 41 8Google Scholar

    [17]

    杨斯雯, 黄玲玲, 王涌天 2023 光学技术 49 522

    Yang S W, Huang L L, Wang Y T 2023 Opt. Tech. 49 522

    [18]

    Ma Y C, Chen S, Ermon S, Lobell David B 2024 Remote Sens. Environ. 301 113924Google Scholar

    [19]

    Fu S Y, Tao S Y, Fan H T, He K, Liu X T, Tao Y L, Zuo J X, Zhang X, Wang Y, Sun Y J 2024 Appl. Energy 353 121991Google Scholar

    [20]

    Chen L C, Zhu Y, Papandreou G, Schroff F, Adam H, 2018 European Conference on Computer Vision Berlin Germany, September 8-14, 2018 p833

    [21]

    Treviso M, Ji T C, Lee J U, van Aken B, Cao Q Q, Ciosici M R, Hassid M, Heafield K, Hooker S, Martins P H, Martins A F T, Milder P, Raffel C, Simpson E, Slonim N, Dodge J, Strubell E, Balasubramanian N, Derczynski L, Gurevych I, Schwartz R 2023 Trans. Assoc. Comput. Linguist. 11 826

    [22]

    Zhu R C, Qiu T S, Wang J F, Sui S, Hao C L, Liu T, Li Y F, Feng M D, Zhang A X, Qiu C W, Qu S B 2021 Nat. Commun. 12 2974Google Scholar

    [23]

    Xu Y, Li F, Gu J Q, Bi Z W, Cao B, Yang Q L, Han J G, Hu Q H, Zhang W L 2024 Adv. Photonics Nexus 3 026002

    [24]

    Qu Y R, Jing L Shen Y C, Qiu M, Soljačić M 2019 ACS Photonics 6 1168Google Scholar

    [25]

    Xu D, Luo Y, Luo J, Pu M B, Zhang Y X, Ha Y L, Luo X G 2021 Opt. Mater. Express 11 1852

    [26]

    Zhang J, Qian C, Fan Z X, Chen J T, Li E P, Jin J M, Chen H S 2022 Adv. Opt. Mater. 10 2200748Google Scholar

    [27]

    张森, 席德科, 李华星 2019 工程热物理学报 40 1058

    Zhang S, Xi D K, Li H X 2019 J. Eng. Thermophys. 40 1058

    [28]

    Jia Y T, Qian C, Fan Z X, Ding Y Z, Wang Z D, Wang D P, Li E P, Zheng B, Cai T, Chen H S 2022 Adv. Funct. Mater. 32 2109331Google Scholar

  • [1] 孙涛, 袁健美. 基于迁移学习的钙钛矿材料带隙预测. 物理学报, doi: 10.7498/aps.72.20231027
    [2] 战庆亮, 白春锦, 葛耀君. 基于时程深度学习的复杂流场流动特性表征方法. 物理学报, doi: 10.7498/aps.71.20221314
    [3] 战庆亮, 葛耀君, 白春锦. 基于深度学习的流场时程特征提取模型. 物理学报, doi: 10.7498/aps.71.20211373
    [4] 张娜, 赵健民, 陈克, 赵俊明, 姜田, 冯一军. 编码超构表面实现双波束独立可重构. 物理学报, doi: 10.7498/aps.70.20210344
    [5] 张瑶, 张云波, 陈立. 基于深度学习的光学表面杂质检测. 物理学报, doi: 10.7498/aps.70.20210403
    [6] 孙雪莹, 刘飞, 段景博, 牛耕田, 邵晓鹏. 基于散斑光场偏振共模抑制性的宽谱散射成像技术. 物理学报, doi: 10.7498/aps.70.20210703
    [7] 雷波, 何兆阳, 张瑞. 基于迁移学习的水下目标定位方法仿真研究. 物理学报, doi: 10.7498/aps.70.20210277
    [8] 夏正德, 宋娜, 刘宾, 潘晋孝, 闫文敏, 邵子惠. 基于字典学习的稠密光场重建算法. 物理学报, doi: 10.7498/aps.69.20191621
    [9] 高强, 李小秋, 周志鹏, 孙磊. 基于分形谐振器的远场超分辨率扫描成像. 物理学报, doi: 10.7498/aps.68.20190620
    [10] 邓俊鸿, 李贵新. 非线性光学超构表面. 物理学报, doi: 10.7498/aps.66.147803
    [11] 白敏, 宣荣喜, 宋建军, 张鹤鸣, 胡辉勇, 舒斌. 压应变Ge/(001)Si1-xGex空穴散射与迁移率模型. 物理学报, doi: 10.7498/aps.64.038501
    [12] 吴晨骏, 程用志, 王文颖, 何博, 龚荣洲. 基于十字形结构的相位梯度超表面设计与雷达散射截面缩减验证. 物理学报, doi: 10.7498/aps.64.164102
    [13] 刘曼, 程传福, 宋洪胜, 滕树云, 刘桂媛. 高斯相关随机表面光散射散斑场相位奇异及其特性的理论研究. 物理学报, doi: 10.7498/aps.58.5376
    [14] 陈春霞, 杜 磊, 何 亮, 胡 瑾, 黄小君, 卫 涛. 金属互连电迁移噪声的分形特征. 物理学报, doi: 10.7498/aps.56.6674
    [15] 刘春香, 程传福, 任晓荣, 刘 曼, 滕树云, 徐至展. 随机表面散射光场的格林函数法与基尔霍夫近似的比较. 物理学报, doi: 10.7498/aps.53.427
    [16] 陈 光. 共形平直时空中电磁场和标量场的解. 物理学报, doi: 10.7498/aps.48.992
    [17] 程传福, 亓东平, 刘德丽, 滕树云. 高斯相关随机表面及其光散射散斑场的模拟产生和光强概率分析. 物理学报, doi: 10.7498/aps.48.1635
    [18] 黄超光. 共形不变标量场在Reissner-Nordsttr?m时空的Boulware 态中的重整化能动张量. 物理学报, doi: 10.7498/aps.42.198
    [19] 赵书城. 可积(度量)Weyl时空中的引力和共形规范场理论. 物理学报, doi: 10.7498/aps.40.849
    [20] 许伯威. Z2对称量子链的共形场理论. 物理学报, doi: 10.7498/aps.40.681
计量
  • 文章访问数:  109
  • PDF下载量:  4
  • 被引次数: 0
出版历程
  • 收稿日期:  2024-08-20
  • 修回日期:  2024-10-29
  • 上网日期:  2024-11-08

/

返回文章
返回