用人工神经网络和多元线性回归预测毛细管气相色谱保留指数

Prediction of Capillary Gas Chromatographic Retention Index Data by Combined Multilinear Regression and Artificial Neural Networks

对石脑油各组分在毛细管气相色谱上的保留指数的理论预测 ,可作为色谱分离过程中识别各组分的一种参考方法 .选择了 6个分子结构信息描述符 ,即分子量、三级分子连接性指数、环烷烃的立体异构体参数及密度 ,作为输入参数以提高 BP网络的预测能力 .对保留指数的同一数据集用多元线性回归和 BP神经网络同时进行了预测 ,平均相对误差分别为 1.2 1%和 1.2 9% .

Neural networks using the back-propagation algorithm have been applied to predict the retention index data of components of naphthas in capillary gas chromatography. Six physicochemical parameters have been set up to describe the structure of compounds. They are molecular weight (M), molecular connectivity indexes( 1 χ, 2 χ, 3 χ), s representing stereoisomers of cycloparaffins and density(d) respectively. Predictions by neural networks are generally in good agreement with the predictions done by multilinear regression techniques. The use of combined statistical method and neural networks can successfully solve the problem of retention index.