机器学习在材料结构与性能预测中的应用综述  

A Review of the Application of Machine Learning in Material Structure and Performance Prediction

作　　者：侯腾跃 孙炎辉 孙舒鹏 肖瑛 郑雁公[2] 王兢[3] 杜海英 吴隽新 HOU Tengyue;SUN Yanhui;SUN Shupeng;XIAO Ying;ZHENG Yangong;WANG Jing;DU Haiying;WU Juanxin(College of Information&Communication Engineering,Dalian Minzu University,Dalian 116600,Liaoning,China;Faculty of Electrical Engineer and Computer Science,Ningbo University,Ningbo 315020,Zhejiang,China;School of Electronic Science and Technology,Dalian University of Technology,Dalian 116024,Liaoning,China;College of Mechanical and Electronic Engineering,Dalian Minzu University,Dalian 116600,Liaoning,China)

机构地区：[1]大连民族大学信息与通信工程学院,辽宁大连116600 [2]宁波大学信息科学与工程学院,浙江宁波315020 [3]大连理工大学电子科学与技术学院,辽宁大连116024 [4]大连民族大学机电工程学院,辽宁大连116600

出　　处：《材料导报》2022年第6期161-172,共12页Materials Reports

基　　金：辽宁省自然科学基金资助项目(20180550634)。

摘　　要：材料是时代进步的重要标志,不同的材料结构具有不同的物理、化学性质,进而影响材料的功能特性。为了实现材料的最优化设计,常采用交叉实验的方法,通过开展大量实验探寻材料的最佳结构。该方法步骤繁琐,无法解决涉及高度非线性或大规模组合过程的复杂问题,也很难揭示材料一些罕见特性。随着计算机科学的发展,机器学习作为一种兼顾开发效率以及开发成本的方法,已经逐渐应用于材料发现、结构分析、性质预测、反向设计等诸多领域,并且在材料学研究中展现出惊人的潜力。然而,机器学习在材料科学中的应用仍存在一些瓶颈。数据集的高效获取、异构型数据集的信息处理、基于轻量化数据集的预测模型建立、材料性能的可靠预测等问题制约着该方向的发展,这些也正是该领域亟需解决的关键问题,同时也是机器学习在材料结构与性能预测中研究的热点与难点。近年来,关于机器学习在材料中应用的论文数量逐年增长,利用机器学习指导新型高性能材料合成的案例也比比皆是。通过采用支持向量机、神经网络等机器学习算法训练数据集来构建模型,以预测材料的结构、吸附特性、电学特性、催化性能、力学特性和热力学特性等材料性能,大大推动了机器学习在材料科学领域的发展,并且已经取得重要突破。为了合理地归纳整理该领域的研究成果,指导后续研究,该综述从应用角度出发,讨论了机器学习在材料结构与性能预测中的数据来源、预测模型以及预测结论等,并对机器学习在未来材料领域中的发展进行了展望。Materials are considered as an important indicator of contemporary progress.Materials with different structures have different physical andchemical properties,which influence their functions.Several experiments have been carried out to achieve materials with optimal designs;c ross-over trial is such a method that it is usually adopted.However,it is quite complicated,because it is not suitable for highly non-linear or large-scale combination processes,and it also makes it challenging to reveal rare attributes of materials.Recent advancements in computer science and machine learning have enabled the development of methods to coordinate the efficiency and cost of development in fields such as materials discovery,structural analysis,property prediction,and reverse design,demonstrating remarkable untapped potential in materials science.Machine learning has its limitations,such as the need to ensure efficient collection of data sets,information processing of heterogeneous data sets,establishment of prediction models based on lightweight data sets,and reliability forecasting of properties of materials.Not only are these problems the key issues in the field that urgently need to be resolved,but they also form the crux of research studies on machine learning for the prediction of structures and properties of materials.Thus,solving these problems can boost the progress of materials science.In recent years,the number of papers on the application of machine learning in material science has increased exponentially.Moreover,several studies have employed machine learning to guide the synthesis of novel materials with superior properties.Vector machines,neural networks,and other machine learning algorithms can be used to form data sets and build models for predicting material properties,such as absorption,electrical properties,catalytic performance,mechanical properties,and thermal performance.These developments have considerably contributed toward the development of materials science and have enabled achievement of major breakt

关 键 词：机器学习 材料结构 吸附特性 电学特性 催化性能 力学特性 热力学特性 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TB34[自动化与计算机技术—控制科学与工程]