基于机器学习的多元化合物带隙预测  

Band gap prediction of multiple compounds based on machine learning

作　　者：董延华[1] 佘安琪 王铭 梁久欣 孙宏宇 DONG Yan-hua;SHE An-qi;WANG Ming;LIANG Jiu-xin;SUN Hong-yu(Colledge of Computer Science,Jilin Normal University,Siping 136000,China)

机构地区：[1]吉林师范大学计算机学院,吉林四平136000

出　　处：《吉林师范大学学报（自然科学版）》2022年第2期119-124,共6页Journal of Jilin Normal University:Natural Science Edition

基　　金：国家教育部科技发展中心项目(2020ITA05017)。

摘　　要：针对使用第一性原理计算获得高精度带隙的方法耗时长、效率低,并且在系统地测量大量材料系统带隙时难度大的弊端,结合机器学习中常用的岭回归、Lasso回归、支持向量回归(SVR)和随机森林算法,提出利用已知物理特征预测多元化合物带隙的机器学习方法.最后通过实验给出了基于成分信息特征集对各种模型的比较结果,分析各方法的性能、优势和不足.通过机器学习方法能够实现精准地预测多元化合物带隙,降低研究成本.To address the disadvantages of using first-principles calculations to obtain high accuracy bandgaps,which are time-consuming,inefficient and difficult when measuring bandgaps of a large number of material systems systematically,a machine learning method for predicting bandgaps of multivariate compounds using known physical features was proposed by combining ridge regression,Lasso regression,support vector regression(SVR)and random forest algorithms,which are commonly used in machine learning.Finally,the results of comparing various models based on component information feature sets were given experimentally to analyse the performance,advantages and shortcomings of each method.The machine learning method enables accurate prediction of multivariate compound band gaps and reduces research costs.

关 键 词：多元化合物 机器学习 带隙 第一性原理 二维材料 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]