神经网络法用于恒酒香气成分性质的研究

Study on the properties of aroma components of Fen-flavor Hengjiu by neural network method

为建构恒酒香气成分色谱保留指数和气味活度值的定量结构-性质相关模型,计算了恒酒香气成分的分子连接性指数和电性距离矢量,优化筛选了连接性指数的0X、1X、2X、3X、4Xpc和电性距离矢量的M3、M9、M14、M32共9种参数,将它们与恒酒香气成分的色谱保留指数进行回归分析,并将其作为神经网络的输入层变量,色谱保留指数作为输出层变量,采用9∶3∶1的神经网络结构,构建了预测能力较好的神经网络模型,总的相关系数R为0.9855,计算的预测值与实验值的平均相对误差为2.59%;同理优化筛选了连接性指数的1X、2X、5Xc和电性距离矢量的M1共4种参数,将它们与恒酒香气成分的气味活度值进行神经网络法研究,采用4∶3∶1的网络结构,所得神经网络模型总的相关系数R为0.9997,气味活度预测值与实验值的平均误差为2.49.结果表明,神经网络模型有良好的预测恒酒香气成分色谱保留指数和气味活度值的能力.可以看出,取代基数量及连接方式是影响恒酒香气成分色谱保留指数和气味活度值大小的主要因素,辛酸乙酯、丁酸乙酯和乙酸乙酯等化合物成分对特征香味有重要贡献.

In order to establish QSPR model for the chromatographic retention index and odor activity value of aroma components of Hengjiu, the molecular connectivity indices and the electronegativity distance vectors of aroma components of Hengjiu were calculated. Nine parameters of the molecular connectivity index 0X,1X,2X,3X,4X pc and the electronegativity distance vectors M 3, M9 , M14 , M 32 were selected. Then, the multi-linear method was applied to analyze the nine parameters and the chromatographic retention index. The nine parameters were used as input layer variables of neural network and the chromatographic retention index was used as output layer variable. Besides, the 9∶3∶1 network structure was adopted and neural network method was used to establish a neural network model with good predictive ability. The total correlation coefficient R was 0.985 5. The mean relative error between the predicted values and the experimental values was 2 .59%. Using the same method, four parameters of the molecular connectivity index 1 X , 2 X , 5 X c and the electronegativity distance vector M 1 were selected, and the neural network method was applied to study the odor activity value of aroma components of Hengjiu. Besides, the 4∶3∶1 network structure was adopted and neural network method was used to establish a neural network model with good predictive ability. The total correlation coefficient R was 0.999 7 . The mean error between the predicted values and the experimental values of odor activity value was 2.49. The results showed that the neural network model had good predictive ability of the chromatographic retention index and odor activity value of aroma components. The amount and the way of connection of substituent groups were the main factors that influenced the chromatographic retention index and odor activity value of aroma components of Hengjiu. The constituents of ethyl octanoate, ethyl butyrate and ethyl acetate had great contributions to the characteristic flavor.