摘要
采用人工神经网络(ANN)建立了36种黑莓果酒香气成分的结构与色谱保留时间之间的定量关系(QSRR)模型。以36种黑莓果酒香气成分的分子连结性指数和电拓扑指数作为输入、保留时间作为输出,采用内外双重验证的办法分析和检验所得模型的稳定性,所构建网络模型的相关系数为0.9993、交叉检验相关系数为0.9949、标准偏差为0.1100、残差绝对值小于0.47,应用于外部预测集,外部预测集相关系数为0.9833。为了便于比较,也采用多元线性回归(MLR)法建立了QSRR模型,模型的相关系数为0.9904、交叉检验相关系数为0.9905、标准偏差为1.4896、残差绝对值小于4.32,外部预测集相关系数为0.8973。结果表明,ANN模型获得了比MLR模型更好的拟合效果。
The model of the quantitative structure retention relationship（QSRR）between 36 kinds of aroma components of blackberry wine and the relative retention time was performed by the artificial neural network（ANN）.The stabilization and generalization ability of the model constructed by the artificial neural network（ANN）method were verified by the inner-external set,when using the molecular connectivity indexes（mχ）and electro topological state indexes（En）of the 36 kinds of aroma components of blackberry wine as the inputs of the neural network and the relative retention time as the outputs of the neural network.For the ANN model,the correlation coefficient was 0.9993,the leave one out crossvalidation regression coefficient was 0.9949,the standard error was 0.1100,the absolute values of residual were less than 0.47 and the correlation coefficient of the test set was 0.9833.In order to make contrast,the QSRR model was set up by multiple linear regressions（MLR）method,which showed that the correlation coefficient was 0.9904,the correlation coefficient of the test set was 0.9905,the standard error was 1.4896,the absolute values of residual were less than 4.32 and the correlation of the test set was 0.8973.The results showed that the performance of neural network was better than that of MLR method.
出处
《分析科学学报》
CAS
CSCD
北大核心
2016年第5期697-700,共4页
Journal of Analytical Science
基金
许昌学院优秀青年骨干教师资助项目
关键词
定量结构色谱保留相关
香气成分
人工神经网络
Quantitative structure retention relationship
Aroma components
Artificial neural network
