南京理工大学学报（自然科学版）
南京理工大學學報（自然科學版）
남경리공대학학보（자연과학판）
JOURNAL OF NANJING UNIVERSITY OF SCIENCE AND TECHNOLOGY
2014年
4期
575-580
,共6页
时静洁%陈利平%陈网桦%杨惠
時靜潔%陳利平%陳網樺%楊惠
시정길%진리평%진망화%양혜
分子形状指数%电性拓扑状态指数%烃类%粘度%定量构效关系%多元线性回归方法%支持向量机方法%预测
分子形狀指數%電性拓撲狀態指數%烴類%粘度%定量構效關繫%多元線性迴歸方法%支持嚮量機方法%預測
분자형상지수%전성탁복상태지수%경류%점도%정량구효관계%다원선성회귀방법%지지향량궤방법%예측
molecular shape indices%electrotopological state indices%hydrocarbons%viscosity%multiple linear regression%quantitative structure-property relationship%support vector method%prediction
基于定量结构性质相关性( Quantitative structure-property relationship,QSPR)原理,以2种Kappa分子形状指数与9种电性拓扑状态指数作为描述符,研究了烃类物质粘度(η)与其分子结构间的内在定量关系。以65种化合物作为样本集,随机选择其中52种作为训练集,剩余13种作为测试集,分别采用多元线性回归方法( Multiple linear regression,MLR)和支持向量机方法( Support vector method,SVM)建立模型进行分析测试。研究结果表明：SVM模型对烃类物质粘度具有很强的预测能力,该方法所得模型在模型拟合和预测能力方面大大优于MLR模型。同时,应用Jackknifed法对SVM模型进行了稳健性检验,进一步说明了SVM对于烃类物质粘度的预测模型的稳定性与可靠性。该研究提供了一种新的预测烃类物质粘度的方法。
基于定量結構性質相關性( Quantitative structure-property relationship,QSPR)原理,以2種Kappa分子形狀指數與9種電性拓撲狀態指數作為描述符,研究瞭烴類物質粘度(η)與其分子結構間的內在定量關繫。以65種化閤物作為樣本集,隨機選擇其中52種作為訓練集,剩餘13種作為測試集,分彆採用多元線性迴歸方法( Multiple linear regression,MLR)和支持嚮量機方法( Support vector method,SVM)建立模型進行分析測試。研究結果錶明：SVM模型對烴類物質粘度具有很彊的預測能力,該方法所得模型在模型擬閤和預測能力方麵大大優于MLR模型。同時,應用Jackknifed法對SVM模型進行瞭穩健性檢驗,進一步說明瞭SVM對于烴類物質粘度的預測模型的穩定性與可靠性。該研究提供瞭一種新的預測烴類物質粘度的方法。
기우정량결구성질상관성( Quantitative structure-property relationship,QSPR)원리,이2충Kappa분자형상지수여9충전성탁복상태지수작위묘술부,연구료경류물질점도(η)여기분자결구간적내재정량관계。이65충화합물작위양본집,수궤선택기중52충작위훈련집,잉여13충작위측시집,분별채용다원선성회귀방법( Multiple linear regression,MLR)화지지향량궤방법( Support vector method,SVM)건립모형진행분석측시。연구결과표명：SVM모형대경류물질점도구유흔강적예측능력,해방법소득모형재모형의합화예측능력방면대대우우MLR모형。동시,응용Jackknifed법대SVM모형진행료은건성검험,진일보설명료SVM대우경류물질점도적예측모형적은정성여가고성。해연구제공료일충신적예측경류물질점도적방법。
The quantitative relationship between viscosities and molecular structures of hydrocarbons is investigated based on the quantitative structure-property relationship( QSPR) principle. The model is constructed by using multiple linear regression ( MLR ) and support vector method ( SVM ) on a dataset that consists of 65 compounds. The dataset is randomly divided into a training set(52)and a testing set(13). Molecular descriptors are considered as inputs to the model,involving two kinds of descriptors ( two Kappa molecular shape indices and nine electropological state indices ) . Results show that in terms of model fitting ability and prediction ability,the SVM model is obviously superior to the MLR model. In order to test the stability and prediction capability of the SVM model, good results are observed in Jackknifed cross validation. This paper provides a new and effective method for predicting viscosities of hydrocarbons.
