设为首页 联系我们 加入收藏

当前位置: 网站首页 期刊分类目录 2025第6期 正文

基于机器学习的钙钛矿氧化物带隙的快速预测

作者:时间:2025-12-28点击数:



全文下载: 202506010.pdf 


文章编号: 1672-6987(2025)06-0075-08 DOI: 10.16351/j.1672-6987.2025.06.010


刘海英1,2 冯建光1 董立峰1,3*1.青岛科技大学 材料科学与工程学院, 山东 青岛 266042;2.潍坊科技学院 材料科学与工程学院, 山东 潍坊 262700;3.哈姆林大学 物理系, 美国 圣保罗 55104)

摘要: 带隙()是决定钙钛矿氧化物功能性的重要参数,但是由于组成和结构上的多样性,对庞大的钙钛矿氧化物家族带隙值进行实验探究或者高精度的理论计算一直是一个巨大的挑战。在这项工作中,采用了基于数据的机器学习技术来实现对钙钛矿氧化物带隙值的快速预测。机器学习模型的构建以实验测量的带隙为目标变量,以PBE计算带隙值、GLLB-SC计算带隙值和成分描述符为特征。结果表明,基于径向基函数的支持向量回归模型表现出了优异的预测性能(测试集上值为0.927,均方根误差仅为0.271 eV);同时,基于递归特征消除方法筛选出了最优的6个特征组合,这为理解带隙预测背后的物理规律提供了依据。最后,实现了对预测空间中2 227个钙钛矿氧化物带隙值的高效预测并进一步进行了稳定钙钛矿氧化物的筛选。


关键词: 钙钛矿氧化物; 带隙预测; 机器学习; 支持向量回归


中图分类号: O 644.1        文献标志码: A


引用格式: 刘海英, 冯建光, 董立峰. 基于机器学习的钙钛矿氧化物带隙的快速预测[J]. 青岛科技大学学报(自然科学版), 2025, 46(6): 75-82.


LIU Haiying, FENG Jianguang, DONG Lifeng. Rapid prediction of band gap values of perovskite oxides based on machine learning[J]. Journal of Qingdao University of Science and Technology(Natural Science Edition), 2025, 46(6): -.

Rapid Prediction of Band Gap Values of Perovskite Oxides Based on Machine Learning

LIU Haiying1,2 FENG Jianguang1 DONG Lifeng1,31.College of Materials Science and Engineering, Qingdao University of Science and Technology, Qingdao 266042,China;

2.School of Materials Science and Engineering, Weifang University of Science and Technology, Weifang 262700,China;3.Department of Physics, Hamline University, Saint Paul 55104, USA)

Abstract: Band gap ( is an important parameter to determine the functionality of perovskite oxides (. However, due to the flexibility of compositions and structures, it has been a great challenge to explore the band gap of a large perovskite oxide family experimentally or calculate the value with high-precision density functional theory. In this work, we use data-based machine learning technology to quickly predict the band gap of perovskite oxides. The construction of machine learning model takes experimentally measured band gap values as the target variable, while using PBE/GLLB-SC calculated band gap values and component descriptors as feature set. The results show that the support vector regression model based on radial basis function exhibits excellent prediction performance with an value of 0.271 and a root mean square error of only 0.271 eV on test set. Meanwhile, this paper screens out the optimal 6 feature combinations based on recursive feature elimination methods, which provides a basis for understanding the physical law about the bandgap prediction. Finally, this paper achieves band gap prediction of 2 227 perovskite oxides efficiently and further screened for stable perovskites oxides.


Key words: perovskite oxides; band gap prediction; machine learning; support vector regression

收稿日期: 2024-12-03

基金项目: 国家自然科学基金项目(52002198).

作者简介: 刘海英 1989—) , 女, 博士研究生.     * 通信联系人.


Copyright © 2011-2017 青岛科技大学学报 (自然科学版)