人工神经网络加速计算材料科学的研究进展  

Artificialneural networks accelerate the research progress of computational materials science

作　　者：赵瑞[1] 石竞琛 刘霏凝 王文杰 ZHAO Rui;SHI Jing-chen;LIU Fei-ning;WANG Wen-jie(College of Mathematics and Computer,Jilin Normal University,Siping 136000,China)

机构地区：[1]吉林师范大学数学与计算机学院,吉林四平136000

出　　处：《吉林师范大学学报（自然科学版）》2023年第1期21-33,共13页Journal of Jilin Normal University:Natural Science Edition

基　　金：国家自然科学基金项目(22078124);吉林省科技发展计划项目(YDZJ202201ZYTS629)。

摘　　要：计算模拟为材料科学带来了第一次计算革命,虽然计算模拟方法可以在不需要实验合成的情况下获得材料的一些关键性能,但大多数计算方法只针对于特定的系统,计算和时间成本很高,并且需要较高的专业基础.作为人工智能的分支之一,人工神经网络算法的应用极大地加速了新材料的发现,可以比计算模拟方法快上一个甚至几个数量级.人工神经网络现已被应用于各种材料性质的预测、高分辨率显微镜图像和衍射图的识别,甚至被嵌入到密度泛函计算中,以代替Kohn-Sham方程,从而缩短计算时间.本文讨论了近5年人工神经网络在材料性质预测方面的不同方法、针对显微镜图像和衍射图的微观结构识别方法以及加快DFT计算的技术特点.Computational simulation brought about the first computational revolution in materials science.Although computational simulation methods can obtain some key properties of materials without experimental synthesis,most computational methods are specific to particular systems.It has high computational and time costs and requires a highly professional base.The application of artificial neural network algorithms,one of the branches of artificial intelligence,has greatly accelerated the discovery of new materials and can be one or even several orders of magnitude faster than computational simulation methods.Artificial neural networks are now being used for the prediction of various material properties,high-resolution microscope images and diffraction map recognition,and even being embedded in density functional calculations in place of the Kohn-Sham equation to reduce the computation time.The paper discussed different approaches to artificial neural networks for material property prediction,microstructure identification methods for microscopy images and diffraction maps,and technical features to speed up DFT calculations in the last five years.

关 键 词：人工神经网络 计算材料科学 性质预测 微观结构识别 计算模拟 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]