论文部分内容阅读
应用CODESSA软件计算300种烃类物质的分子结构描述符,用启发式回归(HM)、最佳多元线性回归(B-MLR)法,以筛选出来的分子描述符建立线性回归模型。用B-MLR法所选4个描述符作为支持向量机(SVM)的输入,建立非线性模型。预测结果表明:所建模型稳健,泛化能力强,预测误差小。非线性模型(R~2=0.9884,RMSE=8.7570)的性能优于线性回归模型(HM:R~2=0.9815,RMSE=11.0653;B-MLR:R~2=0.9814,RMSE=11.1041),预测的效果令人满意。
CODESSA software was used to calculate the molecular structure descriptors of 300 hydrocarbon species. The linear regression model was established by using the heuristic regression (HM), the best multiple linear regression (B-MLR) method and the screened molecular descriptors. The four descriptors selected by B-MLR method are used as input of Support Vector Machine (SVM) to establish a nonlinear model. The prediction results show that the model is robust, the generalization ability is strong and the prediction error is small. The performance of the nonlinear model (R ~ 2 = 0.9884, RMSE = 8.7570) was better than that of the linear regression model (HM: R ~ 2 = 0.9815, RMSE = 11.0653; B-MLR: R ~ 2 = 0.9814, RMSE = 11.1041) The effect is satisfactory.