基于LM-BP神经网络算法饱和醇结构拓扑指数与保留指数相关性的研究

Studies on the Relationship Between Topological Indexes and Retention Indexes for Saturated Alcohols Based on the Levenberg-Marquardt Backpropagation Algorithm of Neural Network

基于LM-BP神经网络算法,建立了饱和醇结构拓扑指数和物理化学性质与在不同固定相上保留指数相关性的人工神经模型。网络的传输函数都是线性的(Purelin函数),隐含层有3个神经元。饱和醇包括带有伯、仲、叔基官能团的直链和支链醇。讨论了隐含层神经元数对神经网络的影响,由19个饱和醇得到的网络适合预测测试醇的精确保留指数。与多元线性回归比较,人工神经网络模型预测结果略优于多元线性回归法。

Based on the Levenberg-Marquardt backpropagation（LM-BP） algorithm of neural network,the model about the relationship between topological indexes and physicochemical properties for the structure of saturated alcohols and their structure retention indexes on different stationary phases were established.The transfer functions of network also are linear（purelin function）.There are three neurons on hidden layer.The saturated alcohols contained branched alcohols with their functional group on a primary,secondary,or tertiary carbon atom.The factors of the number of hidden layer neurons to affect network were discussed.The network obtained with 19 representative alcohols are suitable for predicting accurate retention indexes of the tested alcohols.Compared with multiple linear regression（MLR） method,the model of LM-BP is a little better than MLR in prediction.