基于气相色谱与支持向量机的浓香型白酒基酒等级判断模型研究

Grade judgment model of base liquor of strong-flavor Baijiu based on gas chromatography and support vector machine

以浓香型白酒基酒为研究对象,采用气相色谱技术分析基酒中的挥发性风味物质成分,以特定时间间隔内色谱峰面积为数据点进行提取和计算,结合支持向量机(SVM)建立浓香型基酒质量等级判断模型。以SVM模型预测准确率为评价指标,通过单因素试验和响应面优化试验筛选出构建模型的最佳测试集比例和色谱数据,在最优条件下,依次对SVM类型和核函数进行优化,建立四个级别基酒质量等级判断模型,并设计分级多步预测试验分析不同级别基酒的识别情况。结果表明,当测试集比例为75%,色谱数据分析起始时间为5 min,分析结束时间为19 min,数据点间隔为1 min时,c-多类别支持向量分类的径向基(RBF)核函数所建四个级别基酒分级模型的准确率最高为94.0%。按照前述方法优化后,分级多步试验优化结果表明,三级酒分级准确率为100%,一级与二级基酒分级准确率为97.2%,特级与一级基酒分级准确率为95.1%。该研究同时发现己酸乙酯、乙醇、乳酸乙酯、乙酸乙酯对分级模型的累计方差贡献率可达79.1%,是基酒质量等级划分的重要风味物质。

Taking the base liquor of strong-flavor(Nongxiangxing)Baijiu as research object,the volatile flavor components in the base liquor were determined by gas chromatography.The peak area of chromatographic data points was extracted and calculated at specific time intervals,and the quality grade judgment model of strong-flavor base liquor was established by support vector machine(SVM).With SVM model prediction accuracy as evaluation index,the optimal test set ratio and chromatographic data were selected by single factor tests and response surface optimization.Under the optimal conditions,the SVM type and kernel function were optimized successively,the four levels of base liquor quality grade judgment model were established,and the classification of multi-step prediction test was designed to analyze the recognition of different levels of base liquor.The results showed that under the condition of test set proportion 75%,chromatographic data analysis start time 5 min,end time 19 min,and interval of data points 1 min,the accuracy of the four-level base liquor classification model established by the radial basis function(RBF)kernel function of c-multi-class support vector classification was the highest of 94.0%.After optimization according to the above method,the optimization results of multi-step classification test showed that the accuracy of classification of third grade liquor was 100%,the accuracy of first grade and second grade base wine was 97.2%,and the accuracy of super grade and first grade base liquor was 95.1%.At the same time,it was found that the accumulative variance contribution rate of ethyl caproate,ethanol,ethyl lactate and ethyl acetate to the classification model was up to 79.1%,which were important flavor substances for the quality classification of base liquor.