摘要
从马氏手册中收集了69个烷烃分子,采用ChemOffice软件中的MM2方法优化分子结构,以此为初始构象,运用Gaussian03程序中的RHF/6-31G方法优化结构,得到分子的优势构象,并基于分子图形学技术,计算出7个结构描述子。采用人工神经网络和支持向量机,选定其中60个烷烃分子作样本集,其余9个作预测集,建立分子结构和临界温度的定量关系模型,预测结果的均方根误差与Joback法相近,能够满足工程需要;引入正常沸点数据后,预测结果的均方根误差比Joback法减少37%以上,表明本文提出的方法优点显著。
The molecule structure was first optimized useing MM2 method,a more precise optimization was done using Gaussian03 with the RHF/6-31G method,obtaining the dominant conformation of molecular.Seven structure desccriptors were calculated based on the technology of molecule graphics.This paper chose 60 alkanes as the sample set,using artificial neural network and support vector machine to establish QSPR model to predict the critical temperature,the root-mean-square error of prediction results were close to the group contribution method;When the normal boiling point was introduced,the RMS error of prediction values was reduced over 37%comparing that of Joback method,which indicated that our method is promising.
出处
《计算机与应用化学》
CAS
CSCD
北大核心
2010年第7期979-982,共4页
Computers and Applied Chemistry
关键词
临界温度
描述子
Joback基团贡献法
人工神经网络
支持向量机
critical temperature
descriptor
group contribution method of Joback
artificial neural network
support vector machine
