基于深度学习的纳米粒子阵列电场预测
Prediction of electric field of nanoparticle array based on deep learning method
胡燕萌 1马子轩 1李猛猛1
作者信息
- 1. 南京理工大学 电子工程与光电技术学院,江苏 南京 210094
- 折叠
摘要
纳米粒子阵列超表面的电场计算需要耗费大量的时间和算力.为了实现纳米粒子阵列电场的快速计算,该文借助深度学习方法,提出了一种由低精度求解电场到高精度求解电场的映射深度神经网络.该神经网络可根据低精度的求解电场,预测出高精度求解电场.模型的均方误差为5.6×10-3,平均范数相对误差为 1.5%.数值结果证明该模型实现了对纳米粒子阵列表面电场的快速准确预测,相比已有的研究成果,该文模型误差低了 1 个数量级.该文工作有助于解决纳米粒子阵列的表面增强拉曼散射快速设计问题.
Abstract
The calculation of the electric fields of nanoparticle arrays metasurface require a lot of time and computing power.To achieve the fast calculation of the electric field of nanoparticle arrays,with the help of the deep learning method,this study proposes a mapping deep neural network from low-precision electric field to high-precision electric field.This neural network can predict the electric field with high-precision according to the electric field with low-precision.The mean square error is 5.6×10-3 and the mean norm relative error is 1.5%.The numerical results confirm that the model can fast and accurately predict the electric field on the surface of a nanoparticle array.Compared with the existing research results,the error of this model is lower by an order of magnitude.This work helps to solve the problem of fast design of nanoparticle arrays with surface enhancement of Raman scattering.
关键词
超表面/表面增强拉曼散射/深度学习/纳米粒子阵列Key words
metasurface/surface enhancement of Raman scattering/deep learning/nanoparticle array引用本文复制引用
基金项目
国家自然科学基金(62222108)
国家自然科学基金(61871222)
中央高校基本科研业务费专项(30921011101)
出版年
2024
国家科技期刊平台
NSTL
万方数据