人工神经网络法预测有机物物性的研究──Ⅰ分子描述码与烯烃沸点的预测

Prediction of Phsical Properties of Organic Compounds by Neural Network Ⅰ Molecular Descriptors and Prediction of Bailing Points of Alkenes

讨论了用人工神经网络法预测物性的几类输入参数．用分子描述码作为输入参数预测了烯烃的沸点，烯烃（链）的相关系数ｒ＝０．９９７５，标准差Ｓ＝１．６８７．环烯烃：相关系数ｒ＝０．９９７０，标准差Ｓ＝２．５４８．优于三维连接性指数的计算值．

A number of inputted structural parameters ,which are used to predict physical properties of organic compounds by artificial neural network, were discussed. With the molecular descriptors, the boiling points of alkenes are predicted. In acyclic alkenes: the relational efficiency r=0. 9975,the standard deviation s=1.687;In cyclic alkenes: r=0. 9970,s=2. 548. Above mentioned results are superior to calculating values by using three dimensional connectiviting index.