基于QSPR方法的脂肪醇化合物闪点预测  

Prediction of Flash Point of Fatty Alcohol Based on Quantitative Structure-property Relationship(QSPR)

作　　者：李冀[1] 吴超[1] 

机构地区：[1]中南大学资源与安全工程学院,湖南长沙410083

出　　处：《中国安全科学学报》2012年第6期14-19,共6页China Safety Science Journal

基　　金：国家自然科学基金资助(50974132)

摘　　要：为提高脂肪醇化合物闪点预测精度,提出基于定量结构-性质关系(QSPR)原理的脂肪醇化合物闪点预测方法。应用Dragon软件计算出91种脂肪醇的分子描述符,利用遗传函数算法(GFA)从1 481个描述符中筛选出3个与脂肪醇闪点关系最密切的分子描述符。分别用多元线性回归(MLR)方法和支持向量机(SVM)方法进行建模,并采用内部验证和外部检验的方式对模型的拟合度、预测性等性能进行验证。结果表明:预测集的MLR方法和SVM方法的平均绝对误差(AAE)分别为2.870 K和2.706 K;均方根误差(RMSE)为3.451 K和3.371 K。SVM模型在精度上略优于MLR模型,而MLR模型更为简单和方便。In order to improve the accuracy of predicting flash point of fatty alcohol compounds, a predi- tion method based on the QSPR principle was proposed. Molecular descriptors of 91 kinds of fatty alcohols were calculated with the use of Dragon software and 3 molecular descriptors the most closely relating to the flash points of fatty alcohols were selected from 1 481 descriptors through genetic function approximation （GFA）. The models of the 3 molecular descriptors were built by MLR and SVM respectively, and the de- gree of fitting and predictability were verified using the internal authentication and external validation. Results show that for approach of multivariable linear regression and SVM, the average absolute error （AAE） of the prediction value is 2. 870 K and 2. 706 K respectively, the root mean square error（RMSE） is 3. 451 K and 3. 371 K. The SVM model of flash point exceeds the MLR model in accuracy, while the MLR model is more simple and convenient.

关 键 词：脂肪醇 闪点 定量结构-性质关系(QSPR) 支持向量机(SVM) 多元线性回归(MLR) 预测 

分 类 号：X915.1[环境科学与工程—安全科学]