论文部分内容阅读
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 LM-BP neural network algorithm, an artificial neural model was established to study the relationship between topological index and physicochemical properties of saturated alcohols and retention indices on different stationary phases. The transfer function of the network is linear (Purelin function), the hidden layer has 3 neurons. Saturated alcohols include straight-chain and branched-chain alcohols with primary, secondary and tertiary functional groups. The influence of hidden layer neurons on the neural network is discussed. The network obtained from 19 saturated alcohols is suitable for predicting the exact retention index of tested alcohols. Compared with multivariate linear regression, artificial neural network model prediction results slightly better than multivariate linear regression.