基于集成方法的电子光谱吸收能计算校正模型  

Correction model of calculation set of electron spectral absorption energy based on integration method

作　　者：崔景霞[1] 王亮[1] 王旭[1] 吕佳阳[1] CUI Jingxia;WANG Liang;WANG Xu;L Jiayang(School of Applied Technology,Changchun University of Technology,Changchun 130102,China)

机构地区：[1]长春工业大学应用技术学院,吉林长春130102

出　　处：《长春工业大学学报》2023年第5期474-480,共7页Journal of Changchun University of Technology

基　　金：吉林省高校科技与社科“十二五”科研规划项目(吉教合字[2015]109号)

摘　　要：针对量子化学计算中密度泛函理论在求解较大或复杂分子体系时所出现的计算量巨大且计算精度无法满足研究需求的问题,通过机器学习与量子化学相结合的方式提出一种集成的计算精度校正模型。该模型以多次回归迭代并最终加权平均的形式将GRNN(广义回归神经网络)基学习器集成进AdaBoost模型,并应用于电子光谱吸收能的TDDFT(依时密度泛函理论)计算精度校正。模型结合了神经网络与迭代集成模型的优势,在包含433个有机分子的TDDFT计算数据集上予以实验,并与其他回归模型进行了参照对比。实验结果显示,该集成模型在推荐的STO-3G基组上可将吸收能的TDDFT计算精度MAE与RMSE分别由0.62与0.79降至0.11与0.14。同时,该模型的OECD评价参数R^(2)、Q^(2)与Q^(2)_(cv)值分别为0.97、0.98与0.99,表明该模型具有准确、稳健、高效的特性,在节省计算实验资源的前提下,体现了其在计算精度、拟合优度与稳定性等各方面的优势。In this paper,an integrated computational accuracy correction model is proposed by combining machine learning with quantum chemistry,aiming at the problem that the computational amount is too large and the computational accuracy cannot meet the research requirements when density functional theory is used to solve large or complex molecular systems.In this model,GRNN(generalized regression neural network)based learner is integrated into AdaBoost model in the form of multiple regression iterations and final weighted average,then applied to the calculation accuracy correction of TDDFT(time-dependent density functional theory)of electron spectral absorption energy.Combining the advantages of neural network and iterative integration model,the model was tested on a TDDFT dataset containing 433 organic molecules,and compared with other regression models.The experimental results show that MAE and RMSE of TDDFT can be reduced from 0.62 and 0.79 to 0.11 and 0.14 respectively on the recommended STO-3G base group.At the same time,the values of R^(2),Q^(2)and Q^(2)_(cv)of OECD evaluation parameters of this model also reach 0.97,0.98 and 0.99,respectively,indicating that this model is accurate,robust and efficient,and embodies its advantages in calculation accuracy,goodness of fit and stability on the premise of saving computational experimental resources.

关 键 词：分子吸收能 依时密度泛函理论 集成学习 ADABOOST 广义回归神经网络 

分 类 号：TP311.1[自动化与计算机技术—计算机软件与理论]