【摘要】 应用HyperChem7．0计算与脂肪醇沸点相关的结构参数对119种脂肪醇的沸点做定量结构性质关系(QSPR)研究。在逐步回归算法筛选出影响脂肪醇沸点的分子偶极矩μ、分子最高占有轨道能EHOMO、分子最低空轨道能ELUMO、分子范德华表面积Sg、摩尔折射率Rm、极化率α、分子质量M和疏水参数logP 8个主要结构参数的基础上,采用ε-支持向量机、多元线性回归,以及径向基函数神经网络算法,通过留一法交叉验证建立脂肪醇沸点的QSPR预测模型,3种模型中ε-支持向量机、多元线性回归和径向基函数神经网络模型留一法预测结果的相关系数R分别为0．993、0．988、0．987,标准偏差s则分别为4．774、6．501、6．724,表明ε-支持向量机模型具有最好的预测效果。更多还原

【Abstract】 The quantitative structure property relationships ( QSPR) of boiling point for 119 fatty alcohols is set up with some quantum-chemical parameters calculated by HyperChem7.0. Molecular descriptors, including the dipole moment(μ), the energy of highest occupied molecular orbital (EHOMO) , the energy of lowest unoccupied molecular orbital(ELUMO) ,the Van der Waals surface area(Sg), the refractivity (Rm) , the polarizability(α) , the molecular mass(M) and the log partition coefficient(logP) , are selected by stepwise regression , and the predicting models in this study are developed using cross-validation based onε-support vector machine (ε-SVM), multiple linear regression(MLR) and radical basis function neural network(RBFNN). The correlation coefficient R and standard derivation s gotten by leave-one-out method prediction ofε-SVM, MLR and RBFNN model are 0. 993,0. 988,0.987 and 4. 774,6. 501, 6. 724 respectively. The results show thatε-SVM is the best for predicting accuracy.更多还原