基于聚类分析实现高温大曲发酵过程品温的精准预测  

Accurately predict product temperature during high-temperature Daqu fermentation process based on mathematical modelling

作　　者：姚翠萍 李玉萍 伍中玉 朱楚天 朱治宇 程艳波 焦富 龚佳欣 贾文生 牟明月 黄永光 YAO Cuiping;LI Yuping;WU Zhongyu;ZHU Chutian;ZHU Zhiyu;CHENG Yanbo;JIAO Fu;GONG Jiaxin;JIA Wensheng;MOU Mingyue;HUANG Yongguang(Kweichow Moutai Co.,Ltd.,Zunyi 564500,China;Baijiu Manufacturing Innovation Institute of Guizhou Province,Zunyi 564500,China;School of Mathematics and Statistics,Guiyang 550025,China;School of Brewing and Food Engineering,Guiyang 550025,China)

机构地区：[1]贵州茅台酒股份有限公司,贵州遵义564500 [2]贵州省白酒制造创新研究院,贵州遵义5645003 [3]贵州大学数学与统计学院,贵州贵阳550025 [4]贵州大学酿酒与食品工程学院,贵州贵阳550025

出　　处：《中国酿造》2025年第3期121-128,共8页China Brewing

基　　金：贵州省科技支撑计划(一般项目)(黔科合支撑[2022]一般022);国家自然科学基金项目(32360571,32472319);贵州省科技项目(ZK[2022]047);贵州省信息产业厅项目(黔资发[2020]198号)。

摘　　要：为了科学地评价和预测发酵曲仓整仓大曲的发酵温度变化趋势,该研究选择了春、夏、秋、冬四个季节中具有代表性的制曲轮次(对应第一、三、四、五轮次),对两个代表性制曲车间发酵曲仓的1300余块大曲的发酵温度变化进行全面跟踪采集、数据分析研究。采用聚类分析(CA)和主成分分析(PCA),建立了发酵曲仓整仓大曲制曲品温的发酵温度预测模型。为了进一步验证预测模型的可靠性,选择了另一个制曲发酵车间8个监测区域的曲块温度数据,随机配对并输入模型开展再次验证。结果表明,第一次翻曲阶段模型预测温度与实际温度的总误差<1.2℃,第二次翻曲阶段温度的总误差<1.1℃。预测模型在四个季节制曲温度变化数据的测试中平均准确率>95%,与曲堆整体的实际温度并无显著性差异。该预测模型为有效评价大曲发酵整仓曲块温度变化提供了科学方法,对高温大曲的发酵质量评价具有重要意义。In order to scientifically evaluate and predict the fermentation temperature change trend of Daqu,using the representative Daqu-making rounds(the first,third,fourth and fifth rounds)in spring,summer,autumn and winter as research materials,comprehensively tracked and collected the fermentation temperature changes of 1300 pieces of Daqu in two representative fermentation workshops.The fermentation temperature prediction model for the temperature of Daqu products in the whole workshop was established by principal component analysis(PCA)and cluster analysis(CA).To further verify the reliability of the prediction model,the temperature data of eight monitoring areas in another Daqu fermentation workshop were randomly paired and fed into the model for further verification.The results showed that the total temperature error of the predicted temperature and actual temperature of the first turning stage was less than 1.2℃,and the total temperature error of the second turning stage was less than 1.1℃.The average accuracy of the prediction model of Daqu-making temperature change data in four seasons was more than 95%,and there was no significant difference with the actual model in the test.The prediction model provided a scientific method for effective temperature evaluation of the whole pile,and also provided a scientific basis for temperature control of Daqu fermentation and its intelligent development.

关 键 词：白酒 高温大曲 主成分分析 聚类分析 预测建模 温度监测 

分 类 号：TS262.3[轻工技术与工程—发酵工程]