应用定量结构-性质相关性研究预测液态烃燃烧热

Quantitative Structure-Property Relationship Study for Predicting Heat of Combustion of Liquid Hydrocarbon

分别以基于Xu指数的原子类型AI指数和电性拓扑状态指数作为分子结构描述符表征80个液态烃的分子结构特征,并分别结合人工神经网络和多元线性回归方法,对这80个液态烃的燃烧热进行定量结构-性质相关性建模和预测研究.结果表明,基于Xu指数的原子类型AI指数能更好地表征液态烃物质的分子结构特征,且液态烃燃烧热与分子结构间的线性关系要强于非线性关系.所建立的最佳预测模型为基于Xu指数的原子类型AI指数多元线性回归模型,其模型复相关系数为0.999,对测试集的平均预测相对误差为0.637%,模型预测值与实验值具有较好的一致性.

Both Xu index based atom-type AI indices and electrotopological state indices were used to describe the structures of 80 liquid hydrocarbon molecules, and quantitative structure-property relationship（QSPR） models were developed to predict the heat of combustion of those 80 liquid hydrocarbon by using the artificial neural network and the multilinear regression approach, respectively. The results show that the characteristics of liquid hydrocarbon molecular structures can be better described by Xu index based atom-type AI indices. Furthermore, the linear relationship between the heat of combustion of liquid hydrocarbon and molecular structure is more obvious than the nonlinear relationship. The optimal model was obtained by combining of atom-type AI indices and multi-linear regression, whose correlation coefficient and average relative errors for the testing set were 0.999 and 0. 637% respectively. The predicted values of the models are in good agreement with the experimental data.