支持向量机用于高分子聚合物的折射率预测  

Application of Support Vector Machine to Prediction of Refractive Index for High Molecular Weight Polymers

作　　者：周鹏[1] 梅虎[2] 田菲菲[1] 李志良[1] 

机构地区：[1]重庆大学化学化工学院 [2]生物力学与组织工程教育部重点实验室,重庆400044

出　　处：《应用化学》2006年第12期1410-1412,共3页Chinese Journal of Applied Chemistry

基　　金：化学生物传感与计量学国家重点实验室基金(05-12-1);重庆应用基础研究基金(01-3-6);重庆大学创新基金(06-1-10)资助项目

摘　　要：引入支持向量机(SVM)技术对95个高聚物的折射率进行定量构性相关(QSPR)研究,并将其与多元线性回归(MLR)、神经网络(ANN)进行了比较。结果表明,对于小样本非线性问题支持向量机具较强稳定性及泛化能力,多数情况优于传统方法。所得模型复相关系数r2、交叉检验q2及对外部样本预测能力q2ext分别为0.943、0.938和0.919。Support vector machine （SVM） was employed in quantitative structure-property relationship （QSPR） study for analyzing the refractive indices（n） of 95 high molecular weight polymers. Comparisons was systematically made based on the results obtained with different methods of multiple linear regression（MLR） , artificial neural network（ANN） and ones reported in literatures. It is suggested that SVM possesses both high modeling stability and good generalization ability, especially when applied to investigating both nonlinear systems and small samplings, yielding superior modeling results. The cumulative multiple correlation coefficient r^2 , cumulative cross-validated q^2 and external qext^2 are 0. 943,0. 938 and 0. 919, respectively. Therefore it is believed that SVM has broad prospects and wide applications for the evaluation of polymer properties.

关 键 词：支持向量机 定量构性相关 高聚物 折射率 

分 类 号：O641[理学—物理化学]