基于质谱与化学计量学的白酒原产地鉴定

Identification of Chinese Liquors from Different Geographic Origins Based on Mass Spectrometry and Chemometrics

不同白酒原产地的鉴定对控制白酒质量和保护消费者利益有重要意义。采用顶空同相微萃取与质谱（HSSPMEMS）联用技术获取不同香型和产地的131个白酒酒样在m／z55～19l范围内的离子丰度数据，结合偏最小二乘-判别分析和逐步线性判别分析法筛选m27个重要特征离子，交叉验证的原产地预测准确率达99．2％；然后用筛选出的27个特征离子构建反向传播（BP）神经网络模型和支持向量机（SVM）模型，其原产地预测准确率分别达96．2％和97．7％。其中BP网络的最优参数组合为传递函数logsig、训练函数trainlm、隐藏层神经元数8；而sVM的最优核参数g和惩罚因子c值分别为2和0．125，从参数优化过程及原产地预测准确率可看出，SVM模型对原产地的鉴定效果明显优于BP网络模型。

Determination of the for controlling liquor quality a Chinese liquors from different nd safeguarding the interests o geographic origins is benefit f consumers. In this study, mass spectra of 131 Chinese liquor samples from different geographic origins were collected by the headspace （HS）-solid phase microextraction （SPME）-mass spectrometry （MS）, without squares methods 99.2%. pre-treatment or chromatographic separation. By combination of partial least discriminant analysis （PLS-DA） and stepwise linear discriminant , 27 characteristic ions are finally selected and the predic And then a back-propagation （BP） neural network and tion abili analysis ty of the a support vector （ SLDA ） SI.DA is machine（SVM） recognition model are built, whose prediction accuracy are up to 96. 2% and 97.7%, respectively. The parameter optimization result of BP neural network is logsig, trainlm and eight neurons in hidden layer, while the parameter c, g is 2 and 0. 125 in the SVM model. According to the optimization procedure and prediction accuracy, the SVM model is superior to the BP neural network.