用分子子图对烷烃摩尔响应值的估计与预测

提出了一种新的烷烃拓扑子图表示方法，并结合多元线性回归算法和反传神经网络算法，对烷烃摩尔响应值进行处理，获得了比文献更佳的预测效果，交互校验的相关系数ｒ＝０．９８９。

Prediction and Estimation on Molar Response Values of Alkane Using Molecular Subgraph

A new method based on topographical subgraph of alkanes is proposed and used for molar response values of alkanes. A novel coding, which derived from the numbers of molecular subgraph, is used to characterize molecular structure of alkanes. It showed that there is a very good correlation between the coding and molar response values. The results of multiple linear regression are as follows. Molar response values (S M) on FID: S M=89.579-56.579x 1-40.579x 2-18.191x 3-2.677x 4 11.818x 5 25.690x 6 40.421x 7 54.421x 8 13.926x 9 15.524x 10 13.923x 11 13.857x 12 30.079x 13 30.403x 14 32.677x 15 29.030x 16 , N=50, r= 0.998 3, S=1.934 2; Molar response values (S M) on TCD: S M=139.799 -62.251x 2-49.214x 3-31.244x 4-15.992x 5 0x 6 13.201x 7 27.201x 8 11.152x 9 9.225x 10 8.427 x 11 10.041x 12 22.452x 13 14.977x 14 22.318x 16 , N=32 , r=0.996 3, S=1.766 8 Where r, S and N are regression coefficient, residual standard deviation and data number, respectively. Having tested the modeling by using back\|propagation neural network (BPNN) and leave\|one\|out method, we got the correlation coefficient of cross validation, 0.989.