【作者】 牛晓颖；

【导师】 应义斌；

【作者基本信息】 浙江大学 ， 农业电气化与自动化， 2009， 博士

【摘要】 黄酒是我国的传统酒种,绍兴黄酒集营养和保健为一体,以其低耗粮、低酒度、高营养而备受广大消费者的喜爱,在整个黄酒业的生产、消费以及出口中占有主导地位。随着消费者对于绍兴黄酒中营养成分的关注度不断提高,为黄酒中的风味营养成分寻找一种快速检测方法成为国家检测部门、厂家、商家以及消费者各方面的共同要求。同时,绍兴黄酒陈酿机理的不明确,致使使用近红外进行绍兴黄酒酒龄鉴别的研究难以深入进行下去,因此,对陈酿中绍兴黄酒的多项风味营养成分进行理化和近红外的研究已成为改善黄酒科研现状、提高黄酒消费档次、促进黄酒业规范化、科学化和国际化发展的迫切要求。本研究是对本课题组前期研究(基于傅立叶变换近红外的绍兴黄酒中酒精度、糖度、pH值和总酸指标定量,基酒酒龄1、2、3、4、5年陈定性鉴别)的进一步深入和拓展,目的是实现绍兴黄酒风味营养成分及其基酒酒龄(新酒1年陈、1-3年陈、5年陈和8-10年陈)和勾兑酒酒龄(3年陈、5年陈、6年陈和8年陈)的快速检测和鉴别。本文以绍兴黄酒基酒和勾兑酒为研究对象,利用傅立叶变换近红外光谱分析技术、现代仪器分析技术和化学计量学分析方法,开展绍兴黄酒中总糖、非糖固形物、葡萄糖、异麦芽糖、异麦芽三糖、麦芽糖、潘糖、乙酸、柠檬酸和15种氨基酸的定量检测,基酒及勾兑酒酒龄傅立叶变换近红外分类鉴别及其风味成分变化机理的研究,并在此基础上建立绍兴黄酒各风味成分指标、基酒及勾兑酒酒龄快速鉴别的傅立叶变换近红外光谱分析模型。本文的主要研究内容和研究结论如下:(1)为研究绍兴黄酒陈酿机理及建立绍兴黄酒近红外光谱指纹库,提出了十年不间断、动态研究绍兴黄酒基酒各风味成分指标的样品采集方案,以及将动态实验数据和非动态实验数据相结合研究绍兴黄酒基酒陈酿过程中风味成分和光谱变化的实验方案,设计了用于绍兴黄酒勾兑酒酒龄鉴别研究的勾兑酒配方方法。(2)对绍兴黄酒基酒和勾兑酒样品的近红外原始光谱和预处理光谱特征进行了分析,发现基酒和勾兑酒光谱特征基本相同,均在982、1185、1460、1692、1776、1934以及2265和2302 nm波长附近有比较明显的吸收峰,其中982、1185、1460、1934 nm处的吸收可能与O-H基团有关,而1692、2265和2302 nm处的吸收则可能与C-H基团有关,1776 nm处的吸收可能与糖类成分的吸收有关,各特征峰与绍兴黄酒中糖、酸、氨基酸等风味成分所含有的主要基团C-H、O-H、N-H等的吸收直接相关。(3)分析了绍兴黄酒基酒1年陈、1-3年陈、5年陈和8-10年陈样品中5种糖、3种酸和16种氨基酸的含量变化趋势,发现异麦芽三糖、柠檬酸和蛋氨酸的含量可以作为区别绍兴黄酒新酒1年陈和陈酿后陈年酒(1-3年陈、5年陈和8-10年陈)的判别依据。(4)对使用Chauvenet检验剔除光谱异常样品后的绍兴黄酒基酒新酒1年陈、1-3年陈、5年陈和8-10年陈共63个样品进行了定性分析。结果如下:通过分析7个不同光程(0.5、1.0、1.5、2.0、2.5、3.0和5.0 mm)的分类结果,发现1.0 mm光程分类结果最优。通过比较判别分析(Discriminant Analysis,DA)、判别偏最小二乘(DiscriminantPartial Least Squares,DPLS)、簇类独立软模式分类法(Soft Independent Modeling ofClass Analogy,SIMCA)和最小二乘-支持向量机(Least Squares-Support VectorMachine,LS-SVM)的分类结果,以及全波段、800-1250 nm、1250-1650 nm、1650-2200nm、2200-2500 nm和1250-2200 nm共6个波段和Savitzky-Golay滤波平滑(包括5点、15点和25点三种)、多元散射校正(Multiplictive Scatter Correction,MSC)、标准归一化(Standard Normal Variate,SNV)以及一阶和二阶微分共7种光谱预处理方法对于分类模型的优化效果,发现1.0 mm光程全波段前8个主成分得分输入的LS-SVM方法判别结果最优,校正集正确率97.87%,预测集正确率93.75%,1年陈、1-3年陈和8-10年陈判别正确率为100%,5年陈判别正确率为75%。(5)对剔除光谱异常样品的绍兴黄酒勾兑酒3年陈、5年陈、6年陈和8年陈共99个样品进行了定性分析,结果如下:通过比较DA、DPLS、SIMCA和LS-SVM的分类结果,以及6个建模波段和7种光谱预处理方法对于分类模型的优化效果,发现1.0 mm光程全波段前6个主成分得分输入的LS-SVM方法判别结果最优,校正集和预测集正确率均达到100%。(6)对绍兴黄酒中的总糖、非糖固形物、葡萄糖、异麦芽糖、异麦芽三糖、麦芽糖、潘糖、乙酸、柠檬酸和15种氨基酸进行了定量分析,结果如下:通过比较各指标不同光程的PLSR定量模型,发现总糖、非糖固形物、葡萄糖、麦芽糖、天门冬氨酸、苏氨酸、丝氨酸、甘氨酸、丙氨酸、缬氨酸、异亮氨酸、亮氨酸、酪氨酸、苯丙氨酸、乙酸的0.5 mm光程定量模型效果最优,潘糖2.0 mm光程定量模型效果最优,赖氨酸和精氨酸2.5 mm光程定量模型效果最优,异麦芽糖、异麦芽三糖、柠檬酸、谷氨酸、脯氨酸、组氨酸5.0 mm光程定量模型效果最优。使用杠杆值和学生残差的方法剔除浓度异常样品后用于总糖、非糖固形物、葡萄糖、异麦芽糖、异麦芽三糖、麦芽糖、潘糖、乙酸、柠檬酸定量分析的样品数分别为83、84、123、123、119、120、122、123、124个,用于天门冬氨酸、丝氨酸、谷氨酸、脯氨酸、甘氨酸定量分析的样品数为各90个,丙氨酸为91个,苏氨酸、亮氨酸和组氨酸为各89个,缬氨酸、异亮氨酸、酪氨酸、苯丙氨酸和精氨酸为各88个,赖氨酸87个。使用PLSR、PCR、SMLR和LS-SVM四种定量方法建立各风味成分指标定量模型,并使用6个建模波段和7种光谱预处理方法对各模型进行优化,发现:①使用PLSR方法的1)全波段结合25点平滑处理所建立的总糖、非糖固形物、麦芽糖、天门冬氨酸、甘氨酸、异亮氨酸、苯丙氨酸、脯氨酸定量模型,结合15点平滑处理建立的葡萄糖、潘糖、丝氨酸、丙氨酸、缬氨酸、亮氨酸、赖氨酸定量模型,结合5点平滑建立的苏氨酸模型,结合MSC校正所建立的酪氨酸模型,结合原始光谱所建立的异麦芽三糖、组氨酸模型结果最优;2)800-1250 nm波段结合MSC校正的柠檬酸模型,结合SNV校正的谷氨酸定量模型结果最优。使用SMLR方法的5个波长(958、1182、947、998和989 nm)结合SNV校正的精氨酸定量模型结果最优。使用LS-SVM方法的全波段10主成分得分输入所建立的异麦芽糖、乙酸模型结果最优。②各风味成分指标最优模型的校正集相关系数、交叉验证相关系数、校正误差、预测误差和交叉验证误差分别为总糖0.98677、0.96119、0.741g/L、1.1g/L和1.26g/L,非糖固形物0.98264、0.94269、1.18g/L、1.24g/L和2.12g/L;葡萄糖0.91173、0.79683、1330 mg/L、1550 mg/L和1950 mg/L,异麦芽三糖0.97566、0.84996、223 mg/L、320mg/L和537 mg/L,麦芽糖0.81837、0.699、665 mg/L、748 mg/L和841 mg/L,潘糖0.94010、0.82750、428 mg/L、506 mg/L和706 mg/L;柠檬酸0.98147、0.93231、224mg/L、410 mg/L和423 mg/L;天门冬氨酸0.96745、0.8554、109 mg/L、215 mg/L和226 mg/L,苏氨酸0.96552、0.66033、69.6 mg/L、153 mg/L、202 mg/L,丝氨酸0.90998、0.78218、201 mg/L、290 mg/L、304 mg/L,谷氨酸0.94025、0.8184、281 mg/L、377mg/L、479 mg/L,脯氨酸0.96682、0.70022、244 mg/L、604 mg/L、693 mg/L,甘氨酸0.9416、0.82303、193 mg/L、305 mg/L、328 mg/L,丙氨酸0.98693、0.83257、121mg/L、450 mg/L、417 mg/L,缬氨酸0.9374、0.83179、84.7 mg/L、182 mg/L、136 mg/L,异亮氨酸0.87051、0.71624、66.8 mg/L、75.9 mg/L、96.6 mg/L,亮氨酸0.93946、0.86623、149 mg/L、266 mg/L和217 mg/L,酪氨酸0.97965、0.78817、48.8 mg/L、157 mg/L、150 mg/L,苯丙氨酸0.92269、0.77895、77.8 mg/L、157 mg/L、133 mg/L,赖氨酸0.92269、0.77895、101 mg/L、122 mg/L、167 mg/L,组氨酸0.96504、0.81526、86.9 mg/L、177mg/L、192 mg/L,精氨酸0.78312、0.72322、399 mg/L、592 mg/L和445 mg/L。校正集相关系数、预测集相关系数、校正误差和预测误差分别为异麦芽糖0.91417、0.70894、210 mg/L和264 mg/L,乙酸0.93979、0.79075、157.5 mg/L和283.6 mg/L。使用PLSR载荷光谱提取了24种风味成分的吸收特征波长,但由于各成分所含基团的相似使各成分在近红外区域的吸收重叠在一起,各成分吸收特征波长的特异性还需要进一步的研究确定。各风味成分指标的近红外定量模型精度和稳定性仍需进一步提高,基酒酒龄和勾兑酒酒龄近红外鉴别模型仍需通过增加样本量、样本类型等措施来提高其判别准确率或考察其推广性。本文的研究展示了近红外光谱在快速定量检测绍兴黄酒风味成分以及基酒和勾兑酒酒龄快速鉴别中的潜力,为研发基于近红外光谱的黄酒风味成分快速定量检测和市售黄酒酒龄的快速鉴别仪器打下了良好的基础,并为绍兴黄酒计算机勾兑系统的研制提供了依据。更多还原

【Abstract】 Chinese rice wine is a kind of traditional alcoholic beverage in China. Shaoxing rice wine is the best representation of Chinese rice wine, and it is popular among consumers due to its low consumption of grain, low percent alcohol and high nutritional value. Shaoxing rice wine is in the lead in production and consumption (especially export) of Chinese rice wine. With increasing consumer concern of nutrient components in Shaoxing rice wine, it becomes a demand of quality control departments of government, breweries, sealers and consumers to find a rapid technology or method to detect nutrient components in Shaoxing rice wine. At the same time, because the aging mechanisms is uncertain, the intensive research on discrimination of Shaoxing rice wine age can not be carried out. So making a study of nutrient components in Shaoxing rice wine based on chemical method and near infrared is needed urgently in order to develop Shaoxing rice wine scientific research, improve its quality and grade, accordingly to promote Shaoxing rice wine industry toward standardization, scientization and internationalization. This study is a further extension of the former research projects done by our group members (quantification of alcoholic degree, soluble solids content, pH and total acid; discrimination of Shaoxing rice base wine with age of 1, 2, 3, 4 and 5 years old). The aim of this study is to rapidly detect flavor components, base wine ages (1, 1-3, 5 and 8-10 years old) and blended wine ages (3, 5, 6 and 8 years old) of Shaoxing rice wine.The objects of this research are base wine and blended wine of Shaoxing rice wine. Detection of flavor components were carried out based on fourier transform near infrared spectroscopy (FT-NIR), chemometrics techniques combined with modern physical-chemical analysis techniques. Quantitative models for total sugar, nonsugar solid, glucose, isomaltose, isomaltotriose, maltose, panose, acetic acid, citric acid and fifteen amino acids, were established using NIR spectra and reference data determined by physical-chemical methods. In this dissertation, age classification and discrimination of base wine and blended wine were also studied using FT-NIR spectroscopy technique and pattern recognition methods. Qualitative models were established for classification and discrimination of base wine ages (1, 1-3, 5 and 8-10 years old) and blended wine ages (3, 5, 6 and 8 years old) respectively. The explanation for discriminant results of base wine ages using concentration of flavor components in Shaoxing rice wine samples were carried out too.The main contents and conclusions were:1. The sampling plan of continuously investigating flavor components in Shaoxing rice wine in ten years, and the combination of continuous data and non-continuous data to analyze changing trends of flavor components and NIR spectra of base wine samples, were put forward. The blending formulas of Shaoxing rice wine blended wine samples, which were used to establish qualitative models for discrimination of blended wine with different marked ages, were designed.2. The curve characteristics of original and pretreatment (smooth and derivative) spectra of base wine and blended wine samples were analyzed. Spectra of base wine and blended wine had a very similar shape, and mainly showed absorption bands at around 982, 1185, 1460, 1692, 1776, 1934, 2265 and 2302 nm. The absorption at 982, 1185, 1460, 1934 nm might be related with O-H group; 1692,2265 and 2302 nm with C-H group; 1776 nm with some sugars. The absorption at these bands directly related with flavor components such as sugars, organic acids and amino acids in Shaoxing rice wine.3. Changing trends of five sugars, three acid and sixteen amino acids in Shaoxing rice wine base wine with age of 1, 1-3, 5 and 8-10 years old, were analyzed. Concentrations of isomaltotriose, citric acid and methionine in 1 year old and aged (1-3, 5, 8-10 years old) base wine were absolutely different, with which 1 year old and aged Shaoxing rice wine can be discriminated.4. After spectra outliers being eliminated, sixty-three base wine samples with age of 1, 1-3, 5 and 8-10 years old were analyzed qualitatively:Comparison results of seven optical path-lengths (0.5, 1.0, 1.5, 2.0, 2.5, 3.0 and 5.0 mm) indicated that classification correctness of the model with optical path-length of 1.0 mm was better than models with other optical path-lengths.Comparison results of DA, DPLS, SIMCA, with six modeling bands (800-2500 nm, 800-1250 nm, 1250-1650 nm, 1650-2200 nm, 2200-2500 and 1250-2200 nm) and seven pretreatment methods (5, 15 and 25 points smooth, MSC and SNV), and LS-SVM input by different principal components indicated that classification correctness of the model established using LS-SVM with input of the first eight principal components were optimal. The optimal model had accurate rates of 97.87% for calibration set and 93.75% for prediction set; 100% for base wine samples with age of 1, 1-3 and 8-10 years old, and 75% for samples of 5 years old.5. After spectra outliers being eliminated, ninety-nine blended wine samples with age of 3, 5, 6 and 8 years old were analyzed qualitatively:Comparison results of DA, DPLS, SIMCA, with six modeling bands (800-2500 nm, 800-1250 nm, 1250-1650 nm, 1650-2200 nm, 2200-2500 and 1250-2200 nm) and seven pretreatment methods (5,15 and 25 points smooth, MSC and SNV), and LS-SVM input by different principal components indicated that classification correctness of the model established using LS-SVM with input of the first six principal components were optimal. The optimal model had a accurate rate of 100% for calibration set and prediction set; for blended wine samples with age of 3, 5, 6 and 8 years old.6. After spectra outliers being eliminated, total sugar, nonsugar solid, glucose, isomaltose, isomaltotriose, maltose, panose, acetic acid, citric acid and fifteen amino acids in Shaoxing rice wine samples were analyzed quantitatively.Comparison results of seven optical path-lengths (0.5, 1.0, 1.5, 2.0, 2.5, 3.0 and 5.0 mm) indicated that for total sugar, nonsugar solid, glucose, maltose, aspartic acid, threonine, serine, glycin, alanine, vlaine, isolecine, leucine, tyrosine, phenylalanine and acetic acid, the quantitative results with optical path-length of 0.5 mm were better than those results with other optical path-lengths; for panose was optical path-length of 2.0 mm; for lysine and arginine was 2.5 mm; for isomaltose, isomaltotriose, citric acid, glutamine and proline was 5.0 mm.After concentration outliers being eliminated by leverage and student residual testing, the number of samples used for quantitative analysis of total sugar, nonsugar solid, glucose, isomaltose, isomaltotriose, maltose, panose, acetic acid, citric acid were 83, 84, 123, 123, 119, 120, 122, 123, 124; of aspartic acid, serine, glutamine, proline and glycin were 90 respectively; of alanine were 91; of threonine, leucine and histidine were 89 respectively; of vlaine, isolecine, tyrosine, phenylalanine and arginine were 88; of lysine was 87.Comparison results of PLSR with six modeling bands (800-2500 nm, 800-1250 nm, 1250-1650 nm, 1650-2200 ran, 2200-2500 and 1250-2200 nm) and seven pretreatment methods (5, 15 and 25 points smooth, MSC and SNV), PCR with seven pretreatment methods, SMLR with different wavelength numbers and seven pretreatment methods, and LS-SVM input by different principal components indicated that:(1) The models established using PLSR with 1) 800-2500 nm combined with 25 points smooth for total sugar, nonsugar solid, maltose, aspartic acid, glycin, isolecine, phenylalanine and proline; combined with 15 points smooth for glucose, panose, serine, alanine, vlaine, leucine and lysine; combined with 5 points smooth for threonine; combined with MSC for tyrosine; combined with original spectra for isomaltotriose and histidine; 2) 800-1250 nm combined with MSC for citric acid and with SNV for glutamine, were optimal.(2) The models established using SMLR with five wavelengths (958, 1182, 947, 998 and 989 nm) combined with SNV for arginine, were optimal.(3) The models established using LS-SVM with input by the first ten principal components for isomaltose and acetic acid, respectively, were optimal.The parameters, the correlation coefficient of calibration set and cross-validation, the root mean square error of calibration, prediction and cross-validation, of the optimal models for twenty-six flavor components were as follows: for total sugar, 0.98677, 0.96119, 0.741 g/L, 1.1 g/L and 1.26 g/L; for nonsugar solid, 0.98264, 0.94269, 1.18 g/L, 1.24 g/L and 2.12 g/L; for glucose, 0.91173, 0.79683, 1330 mg/L, 1550 mg/L and 1950 mg/L; for isomaltotriose, 0.97566, 0.84996, 223 mg/L, 320 mg/L and 537 mg/L; for maltose, 0.81837, 0.699, 665 mg/L, 748 mg/L and 841 mg/L; for panose, 0.94010, 0.82750, 428 mg/L, 506 mg/L and 706 mg/L; for tartaric acid, 0.87063, 0.81603, 21.8 mg/L, 34.6 mg/L and 25.7 mg/L; for citric acid, 0.98147, 0.93231, 224 mg/L, 410 mg/L and 423 mg/L; for aspartic acid, 0.96745, 0.8554, 109 mg/L, 215 mg/L and 226 mg/L; for threonine, 0.96552, 0.66033, 69.6 mg/L, 153 mg/L and 202 mg/L; for serine, 0.90998, 0.78218, 201 mg/L, 290 mg/L and 304 mg/L; for glutamine, 0.94025, 0.8184, 281 mg/L, 377 mg/L and 479 mg/L; for proline, 0.96682, 0.70022, 244 mg/L, 604 mg/L and 693 mg/L; for glycin, 0.9416, 0.82303, 193 mg/L, 305 mg/L and 328 mg/L; for alanine, 0.98693, 0.83257, 121 mg/L, 450 mg/L and 417 mg/L; for vlaine, 0.9374, 0.83179, 84.7 mg/L, 182 mg/L and 136 mg/L; for isolecine, 0.87051, 0.71624, 66.8 mg/L, 75.9 mg/L and 96.6 mg/L; for leucine, 0.93946, 0.86623, 149 mg/L, 266 mg/L and 217 mg/L; for tyrosine, 0.97965, 0.78817, 48.8 mg/L, 157 mg/L and 150 mg/L; for phenylalanine, 0.92269, 0.77895, 77.8 mg/L, 157 mg/L and 133 mg/L; for lysine, 0.92269, 0.77895, 101 mg/L, 122 mg/L and 167 mg/L; for histidine, 0.96504, 0.81526, 86.9 mg/L, 177 mg/L and 192 mg/L; for arginine, 0.78312, 0.72322, 399 mg/L, 592 mg/L and 445 mg/L; the correlation coefficient of calibration and prediction set, the root mean square error of calibration and prediction, for isomaltose, 0.91417, 0.70894, 210 mg/L and 264 mg/L; for acetic acid, 0.93979, 0.79075, 157.5 mg/L and 283.6 mg/L.PLSR loading spectra on the full spectra were investigated to extract absorption features of twenty-four flavor components in Shaoxing rice wine samples. The absorption of these components overlapped due to the structure similarity, so the speciality of the absorption features should be confirmed in future studies. The precision and robustness of quantitative models for flavor components need improved, and the accurate rate and adaptability of qualitative models respectively for base wine and blended wine with different ages should be promoted through adding samples and modeling methods. The results obtained in this study indicated the potentiality of NIR spectroscopy technique to rapidly detect flavor components in Shaoxing rice wine, and discriminate base wine and blended wine with different ages. The study laid foundation for developing an instrument for rapid detecting flavor components and ages of commercial Shaoxing rice wine, and provided bases for establishing computer blending system of Shaoxing rice wine.更多还原