【作者】 李东亮；

【导师】 许自成；

【作者基本信息】 河南农业大学 ， 烟草学， 2008， 博士

【摘要】 烟草是最重要的经济作物,它从一开始就在农业经济和国际贸易中占有重要地位,是国民经济的重要组成部分。烟叶质量评价是风格特色烟叶研究、烟叶加工特性研究、模块配方手段研究和叶组替代技术研究等众多烟草前沿研究方向的基础。本试验以灰色理论为基础,结合神经网络理论、模糊理论等综合评价方法对烤烟化学成分、烤烟产区、烤烟形态特征与化学成分、烤烟感官质量等诸多问题进行了研究和评价,主要研究结论如下:1、众多的烟草化学成分指标增加了烟叶品质分析和质量综合评价的难度。针对这一问题,本试验选取主要烤烟产区样品共247份,并运用灰色统计分析法研究了这些样品的10个化学成分指标。结果表明:氮碱比属于大影响指标,对烟叶品质的影响程度最大;总氮、糖碱比、氯、还原糖、烟碱、水溶性总糖、总植物碱、钾氯比为中影响指标,对烟叶品质的影响程度仅次于大影响指标,蛋白质属于小影响指标。配方叶组选择烟叶原料时,如果烟叶化学成分不在适宜区域内,总氮等大影响指标选择小于适宜区域的化学成分比较合理;还原糖、烟碱、水溶性总糖、总植物碱、钾氯比、糖碱比、氮碱比、氯等中影响指标应选择大于适宜区域的化学成分对烟叶品质损害较小。2、烤烟化学成分指标之间的关系密切而复杂。为简化化学成分指标,本文选取我国主要烟区272份烟叶样品,对14个化学成分指标进行了灰色关联聚类分析。结果表明:当聚类临界值为0.8时,14个化学成分指标可被聚为5类,总氮、糖碱比、氮碱比、氯、总挥发性酸可以分别作为每类的代表;当聚类临界值为0.7时,化学成分指标被聚为酸性和碱性两大类,代表性指标分别为糖碱比和氮碱比。与系统聚类相比较,灰色关联聚类更具有合理性。3、单料烟感官质量评定大多采用描述性语言对各单项指标进行评价,不易形成定量的综合评价结果。针对这一问题,本研究采用层次模糊法对单料烟感官质量进行了综合评价。结果表明:层次分析法确定各单项指标的权重符合实际的评吸要求,模糊综合评价容易给出定量的综合评价结果,在此基础上计算f函数值便于对不同的单料烟样品进行排序。4、为了探索烤烟等级替代、分组加工的理论基础,本试验采用灰色等权聚类法对18个烤烟产区的68份下部叶化学成分进行了聚类,结果表明,福建将乐县下部叶化学成分属于A类,为化学成分很协调区域,其他17个烤烟产区下部叶化学成分都属于B类,为化学成分协调区域。当某个聚类系数与其他聚类系数同时差异显著,且该聚类系数为最大时,综合聚类系数和一般聚类系数结果一致。采用总氮、氮碱比、糖碱比、氯和总挥发性酸5个化学成分指标与采用14个指标对烤烟产区进行灰色等权聚类的结果具有较好的相似性。采用平均值—标准偏差法确定白化函数的值域,能有效避免人为因素对聚类结果的影响。5、为了探讨不同烤烟产区化学成分的相似性,为烤烟分组加工和等级替代提供理论依据,本试验采用因子分析与广义灰色绝对关联相结合的方法对不同烤烟产区烟叶上部叶化学成分进行了统计分析,结果表明,与福建邵武产区烤烟上部叶化学成分最接近的地区为云南宜良,其次是广东南雄,化学成分差异最大的是云南华宁;因子分析法可以解决烤烟化学成分评价指标的多重共线性问题;以灰色绝对关联度为依据,结合烤烟感官质量评定可以评价不同烤烟产区化学成分的相似性;正、负相关与关联度大小的关系对于评价烤烟产区化学成分的接近程度来说没有影响,但负的原始数据需进行转换。6、本试验采用多元方差分析法对不同部位烟叶的化学成分在烤烟产区间的差异性、同一产区不同品种间化学成分的差异性以及不同产区不同品种间化学成分的差异性进行了系统分析,结果表明,各个部位烟叶在各个产区之间的综合效应有极显著的差异;云南省各个部位烟叶在各个品种之间的综合效应有极显著的差异;烤烟B₂F品种之间、产区之间以及品种和产区的交互作用之间的综合效应差异都达到了极显著水平;烤烟C₂F品种之间、品种和产区的交互作用之间的综合效应差异都不显著,产区之间的综合效应差异达到了极显著水平;烤烟C₃F品种之间、产区之间以及品种和产区的交互作用之间的综合效应差异都不显著;烤烟X₂F品种之间、产区之间的综合效应差异都达到了极显著水平,品种和产区的交互作用之间综合效应差异不显著。多元方差分析法可以对多组样本的多指标样本进行综合效应分析,得到多组样本的综合检验结果。该评价结果对烟叶等级替代、分组加工、小配方模块等特色工艺技术更具有实际指导意义。7、本试验采用多因素方差分析法对10个烤烟产区4个年份4个等级的烟叶进行了系统分析,结果表明,不同等级、不同产地和不同年份对烤烟糖碱比的影响大小分别为等级〉产地〉年份,F值和显著水平表明,不同等级、不同产地对烤烟糖碱比的影响都达到了极显著水平;不同年份对烤烟糖碱比的影响不显著。年份和产地、年份和等级的交互作用（共同作用）没有对糖碱比造成显著影响,等级和产地的交互作用对糖碱比造成了显著影响。不同等级、不同产地和不同年份对烤烟氮碱比的影响大小分别为等级)产地)年份,F值和显著水平表明,不同等级、不同产地、不同年份对烤烟氮碱比的影响都达到了极显著水平;年份和等级的交互作用没有对氮碱比造成显著影响,年份和产地、等级和产地的交互作用对氮碱比造成了显著影响。多因素方差分析法可以对产区、等级、年份及其交互作用对糖碱比、氮碱比的影响程度进行有效评价,该评价结果有助于等级替代时各影响因素的取舍。8、以我国主产烟区的主栽烤烟品种K326为材料,测定了3个等级（C₃F、B₂F和X₂F）共计391份样品的烟碱含量和物理性状,分析了烟碱含量与物理性状的相互关系。结果表明:烟碱含量和物理性状在样品间存在着广泛的变异;在一定范围内,烟碱含量与上部叶单叶重呈显著正相关,与上、下部叶长分别呈极显著正相关和显著负相关;与中、下部叶宽分别呈显著负相关和极显著负相关,与厚度和叶面密度均不相关;烟叶烟碱含量与上部叶单叶重的关系可用S形曲线（?）:e^{（1.484-2.041/x）}来描述,与上、下部叶长的关系可分别用逆函数（?）=8.225-265.750/x和（?）=0.427+105.053/x来描述,与中、下部叶宽的关系可分别用二次函数（?）=0.533+0.316x-0.010x²和（?）=-0.157+0.307x-0.009x²来描述,且均达到显著水平,烟碱含量与其他物理性状的曲线回归不显著,表明曲线回归一般较简单线性回归分析更有效。简单相关、曲线回归和次数分布三种方法结合可以较好地反映烟碱含量和物理性状的内在关系。9、本试验以我国主产烟区的主栽烤烟品种K326为材料,测定了X₂F（下橘二）等级共计124份样品的总氮、烟碱含量和物理性状,系统地分析了总氮含量、氮碱比与物理性状的相互关系。结果表明:总氮含量、氮碱比和物理性状在样品间存在着广泛的变异。总氮含量随单叶重、叶长、叶面密度的增加而逐步下降;氮碱比随叶长、叶宽的增加而上升,随叶面密度的增加而逐渐下降。总氮与叶长、叶面密度的关系可分别用逆函数（?）=0.827+38.047/x、复合函数（?）=1.880×0.996^x描述;氮碱比与叶宽、叶面密度的关系可分别用复合函数（?）=0.413×1.023^x描述、幂函数（?）=2.791×x^-0.354描述。10、本试验以我国主产烟区124份烤烟中部叶样品为研究材料,采用数理统计方法系统地分析了还原糖含量、还原糖烟碱比与若干形态特征的相互关系。结果表明:还原糖含量和形态特征在样本间存在着广泛的变异;还原糖、还原糖烟碱比与单叶重的变化趋势相似,即总体上随单叶重增加而呈下降趋势,至单叶重为12.3g时达到最低点,然后又呈上升趋势;随叶长的变化规律也相似,都是随叶长的增加而增大,在叶长超过64.9cm后,二者迅速增加;与叶厚的变化曲线一致,都是随叶厚的增加而逐渐下降,当叶厚大于0.126mm后,还原糖烟碱比下降幅度较大;还原糖烟碱比随叶宽增加逐渐上升;分别建立了单一形态特征对还原糖含量、还原糖烟碱比的曲线回归方程,还原糖与单叶重、叶厚的关系可分别用二次函数（?）=45.243-3.667x+0.139x²、（?）=11.217+316.779x-2026.595x²进行描述,还原糖烟碱比与单叶重、叶厚的关系可分别用对数函数（?）=16.278-3.375ln x、复合函数（?）=14.075×0.002^x进行描述;且均达到5%显著水平,表明曲线回归较简单线性回归分析更有效。经过数据验证,建立的曲线回归方程拟合性较好。11、本研究以我国烤烟主产区122份烟叶中部叶样品为基础,系统地分析了烟叶钾含量、氯含量、钾氯比与物理性状之间的关系。结果表明:各性状在供试材料间存在着广泛的变异。烟叶钾含量与叶长、叶面密度分别呈极显著正相关、负相关;烟叶氯含量与叶面密度呈显著正相关;烟叶钾氯比与叶长、叶厚分别呈极显著正相关、显著负相关。南方烟区烤烟中部叶烟叶钾含量与叶长、叶面密度的关系可分别用逆函数（?）=7.803-343.262/x、复合函数（?）=4.612×0.991^x进行描述;烟叶氯含量与叶面密度的关系可用二次函数（?）=1.268-0.034x+0.0003x²描述;烟叶钾氯比与叶长、叶厚的关系可分别用复合函数（?）=0.023×1.105^x、二次函数（?）=94.408-1355.208x+5897.907x²进行描述;且均达到5%显著水平,表明曲线回归一般较简单线性回归分析更有效。烤烟中部叶钾含量、钾氯比随单叶重的增加而降低,随叶长的增长而增加明显;氯含量随单叶重的增加而升高,随叶长的增加而迅速下降,随叶厚的增加而增加;南方烟区烤烟叶宽为23.5cm、叶面密度为85.31 g/m²是两个敏感闽值。12、本文以我国南方烟区烤烟品种K326不同部位的166份样品为例,研究了烤烟主要化学成分与叶片长度的关系。结果表明,烟碱、总氮、还原糖、钾、氯、氮碱比、糖碱比、钾氯比、叶长在样品间存在广泛的变异。氯和钾氯比在部位间差异不显著,叶长、烟碱、氮碱比和糖碱比在部位间差异显著;总氮、还原糖在中部、下部间差异未达到显著水平,但二者均与上部叶差异显著;上部、中部叶的钾含量差异不显著,但二者均与下部叶差异显著。上部、下部叶长达到55.3cm时比较敏感,超过该长度时,烟碱、总氮、氮碱比出现明显变化;在中、下部叶长大于55.3cm后,还原糖含量也出现突变。上部叶的糖碱比与中、下部叶的钾氯比、钾、氯含量在叶长大于61.5cm后变化较大。13、以烤烟品种K326为供试材料,采用典型相关分析方法研究了烤烟不同部位烟叶的主要物理性状和化学成分的关系。结果表明,烤烟物理性状和化学成分所具有的显著性指标因部位而异;上、中、下3个部位物理性状共有的显著性指标是叶长、叶厚和叶面密度,化学成分共有的显著性指标是氯含量、钾含量和钾氯比,且后者对典型变量的相对作用大小一致,表现为钾＞钾氯比＞氯;烤烟不同部位烟叶叶长、叶厚和叶面密度与钾、钾氯比和氯的关系最为密切。典型负载系数可以作为确定各部位物理特性和化学成分显著性指标的重要依据。14、为了指导配方打叶的小配方模块设计,本试验利用灰色理论建立了烤烟醇化过程糖碱比GM（1,1）灰色模型、糖碱比新陈代谢GM（1,1）灰色模型、氮碱比GM（1,1）灰色模型、氮碱比指数平滑GM（1,1）灰色修正模型和氮碱比对数GM（1,1）灰色修正模型。模型检验结果表明,建立的各种糖碱比、氮碱比模型都有意义,而且可用作中、长期预测。糖碱比灰色模型精度较高,两种模型精度分别达到了98.37%和99.73%;氮碱比灰色模型通过指数平滑和对数修正后,模型精度分别达到了93.56%和88.52%。15、利用灰色理论建立了烤烟醇化期间总香味成分、香气质、香气量的GM（1,1）灰色模型。河南B₂F的总香味成分、香气质、香气量的GM（1,1）灰色模型精度分别达到了98.63%、99.24%和99.91%;河南C₃F的总香味成分、香气质、香气量的GM（1,1）灰色模型精度分别达到了97.82%、99.94%和99.09%。16、建立模型的首要问题是恰当地选取进入模型的解释变量。针对这一问题,本文采用灰色绝对关联度和灰色关联聚类相结合的综合评价方法,对烤烟主要化学成分和香气质的灰色模型变量进行了分析。结果表明,进入香气质灰色预测模型的化学成分指标变量为总氮、还原糖、pH值、糖碱比、氮碱比;进入香气量灰色预测模型的化学成分指标变量为总氮、还原糖、钾、pH值和糖碱比。灰色绝对关联度和灰色关联聚类相结合不仅解决了自变量与因变量之间的关系问题,也解决了自变量之间的多重共线性问题。17、建立了烤烟化学成分与香气质、香气量的GM（1,N）-BP神经网络预测模型,结果表明,香气质GM（1,N）-BP神经网络预测模型的模拟精度、预测精度分别达到95.29%、98.56%,比单纯GM（1,N）模型的预测精度提高了3.4%;香气量GM（1,N）-BP神经网络预测模型的模拟精度、预测精度分别达到95.13%、98.20%,比单纯GM（1,N）模型的预测精度提高了2.16%。当网络结构为1-3-1时,香气质GM（1,N）-BP神经网络预测模型的预测精度最高;当网络结构为1-9-1时,香气量GM（1,N）-BP神经网络预测模型的预测精度最高。18、采用时间序列分析法,建立了卷烟烟气焦油量的预测模型。结果表明,所建立的卷烟烟气焦油量的ARMA（2,2）预测模型,模型预测精度达到了99.51%,平均相对误差为0.49%,属于一级（优等）模型,预测精度优于因果模型;利用时间序列建立预测模型可以避开模型自变量的选择这一难题,但时间序列一般用于短期预测,不能用于长期预测。更多还原

【Abstract】 Tobacco is important economic crop and has great influence on national economy.Tobacco quality evaluation is the base of grades of tobacco leaf substitution, grouping process and little formulation module.The test got such as follows results of tobacco chemical components,the relationship between chemical components and shape character and sensory quality by integrating grey theory,neural network theory and fuzzy theory.1.It was difficult for fiue-cured tobacco leaves to analyze and evaluate quality because of numerous chemical compositions index.To solve the problem,the test analyzed ten routine chemical compositions index of 247 samples from main tobacco-growing areas of China by grey statistic analysis.The results indicated that the great influence index were ratio of total nitrogen to nicotine（RNN） which had the biggest influence on quality of flue-cured tobacco leaves,on the opposite,it was inferior to the great influence index that the influence of total nitrogen（TN）,ratio of total sugar to nicotine（RSN）,chlorine,reducing sugar,nicotine,total sugar,total alkaloid and ratio of potassium to chlorine（RPC） on quality of flue-cured tobacco leaves,and protein was little influence index.If the chemical compositions were beyond optimum range when blending leaf groups selected flue-cured tobacco leaves,it was rational to select less than optimum range for the great influence index such as RNN.On the opposite,it was rational to select more than optimum range for the chemical compositions index such as reducing sugar,nicotine,total sugar,total alkaloid,RPC,TN,RSN and chlorine.2.There existed close and complicated relationship among chemical component indexes in flue-cured tobacco leaves.To simplify chemical component indexes,272 samples from main tobacco-growing areas of China were selected as materials,and 14 common chemical component indexes were analyzed by grey relational degree cluster.The results indicated that the 14 common chemical component indexes could be classified into five groups when cluster critical value was fixed at the level of 0.8.Meantime,total nitrogen,ratio of total sugar to nicotine,ratio of total nitrogen to nicotine,chlorine and total volatile acids were determined the representative indexes for each group.When cluster critical value was fixed at the level of 0.7, the chemical component indexes could be classified into two groups,named acids group and alkaloid group with representative indexes of the ratio of total sugar to nicotine and ratio of total nitrogen to nicotine,respectively.Grey relational degree cluster analysis was more rational than hierarchical cluster analysis.3.It is difficult to obtain mensurable comprehensive evaluation results because descriptive language for different single index was mostly used to evaluating unblended cigarette sensory quality.In order to solve the problem,the hierarchical fuzzy systems were introduced to comprehensive evaluation for unblended cigarette sensory quality.The results indicated that the weight determined by analytic hierarchy process was in accord with actual needs;fuzzy evaluation could easily make mensurable comprehensive results;ffunction values estimated on the basis of analytic hierarchy process and fuzzy transformation were convenient for giving the quality order of different unblended cigarette samples.4.In order to explore the theory basis of flue-cured tobacco grade substitute and grouping process,the test clustered 68 chemical component samples of 18 flue-cured tobacco-growing areas by grey equal weight cluster method.The results indicated that the chemical components of Jiangle county of Fujian province should be clustered A class which belonged to better region of chemical components,and other 17 flue-cured tobacco-growing areas should be clustered B class which belonged to good region of chemical components.The result of comprehensive cluster coefficient and common cluster coefficient was consistent when some cluster coefficient was distinguished difference with other cluster coefficients and it was the biggest.The cluster result was similar by applying 5 indexes including total nitrogen,the ratio of total nitrogen to nicotine,total sugar to nicotine,chlorine and volatile acid and 14 indexes.It could be avoided of influence on cluster results of factitious factor by applying mean-standard deviation method to confirm the values of albino function.5.In order to discuss the chemical components comparability of different main tobacco-growing areas and explore the theory basis of flue-cured tobacco grade substitute and grouping process,the test analyzed the chemical components of different main tobacco-growing areas upper leaves by combining factor analysis and absolute degree of grey incidence.The results indicated the most similar area was Yiliang of Yunnan province with chemical components of upper leaf of Shaowu of Fujian province,the next was Nanxiong of Guangdong province,and the most difference of chemical components was Huaning of Yunnan province.The factor analysis method could solve the problem of multiple mutual linear of chemical components evaluation indexes.On the basis of absolute degree of grey incidence,the chemical components comparability of different main tobacco-growing areas was evaluated combining sensory evaluation.The relation between positive correlation, negative correlation and degree of grey incidence had no effect on evaluation of the chemical components comparability of different main tobacco-growing areas;however,the minus original data were converted.6.The test systems analyzed the chemical components difference of different position tobacco leaves among main tobacco-growing areas,different variety in the same tobacco-growing area and different variety in the different tobacco-growing area,respectively. The results indicated that the comprehensive effect difference was significant at 1%level among main tobacco-growing areas;the comprehensive effect difference was significant at 1% level among different variety at Yunnan province;the comprehensive effect difference was significant at 1%level among different variety,different main tobacco-growing areas and interaction of tobacco-growing areas and variety for B₂F;the comprehensive effect difference was not significant at 5%level among different variety and interaction of tobacco-growing areas and variety,but the comprehensive effect difference was significant at 1%level among different tobacco-growing areas for C₂F;the comprehensive effect difference was not significant at 5%level among different variety,different main tobacco-growing areas and interaction of tobacco-growing areas and variety for C₃F;the comprehensive effect difference was significant at 1%level among different variety and tobacco-growing areas,but the comprehensive effect difference was not significant at 5%level among interaction of tobacco-growing areas and variety for X₂F.The multianalysis of variance could be used the comprehensive effect analysis for more indexes of samples,and the evaluation results had factual significance for grades of tobacco leaf substitution,grouping process and little formulation module.7.The test analyzed tobacco leaves of four grades among four years of ten tobacco-growing areas by multifactor variance method.The results indicated that the influence orders was grades＞tobacco-growing areas＞years on the ratio of total sugar to nicotine of different grades,different tobacco-growing areas and different years,and the influence of different grades and different tobacco-growing areas were the significant at 1%level,but it was not significant at 5%level of different years.The mutual influence of years and tobacco-growing areas,years and grades was not significant at 5%level on the ratio of total sugar to nicotine,but the mutual influence of grades and tobacco-growing areas was significant at 5%level on the ratio of total sugar to nicotine.The influence orders was grades＞tobacco-growing areas＞years on the ratio of total nitrogen to nicotine of different grades, different tobacco-growing areas and different years,and the influence of different grades, different years and different tobacco-growing areas were the significant at 1%level.The mutual influence of years and grades was not significant at 5%level on the ratio of total nitrogen to nicotine,but the mutual influence of grades and tobacco-growing areas,grades and tobacco-growing areas was significant at 5%level on the ratio of total s nitrogen to nicotine. The multifactor variance method could be used the comprehensive evaluation on the influence of tobacco-growing areas,grades and years on the ratio of total sugar to nicotine and the ratio of total nitrogen to nicotine,and the evaluation results had factual significance for grades of tobacco leaf substitution.8.K326,a flue-cured tobacco cultivar widely used in main tobacco-growing areas of China,was selected as materials to determine the nicotine content and physical properties;total 391 samples including three grades（C₃F,B₂F and X₂F） were employed to studying the relationship between the nicotine content and physical properties.The results indicated that there existed extensive variation among samples for physical properties and the nicotine content.In the given range,the nicotine content had significant positive correlation with the single leaf weight of upper leaf,high positive correlation and negative correlation with leaf length of upper leaf and lower leaf respectively,negative correlation and high negative correlation with leaf width of cutters and lower leaf respectively,and no correlation with other physical properties.The relationship could be described by S equation（?）=e^{（1.484-2.041/x）} between nicotine content and single leaf weight of upper leaf,by inverse equation （?）=8.225-265.750/x and（?）=0.427 + 105.053/x between nicotine content and leaf length of upper leaf and lower leaf respectively,by quadratic equation（?）=0.533+0.316x-0.010x² and （?）=-0.157 + 0.307x-0.009x² between nicotine content and leaf width of cutters and lower leaf respectively.All above equations were significant at 5%level.However,the curve regression equation between nicotine content and other physical properties was not significant at 5%level, and were more efficient than simple linear regression.The relationship between nicotine and physical properties could be described by using the methods of linear correlation,curve regression and degree distribution.9.Total 124 samples of X₂F grades of K326,a flue-cured tobacco cultivar widely planted in main tobacco-growing areas of China,were selected as materials to determine total nitrogen content,nicotine content and physical properties;relationships of total nitrogen content and ratio of total nitrogen to nicotine with physical properties were analyzed.The results indicated that there existed largely variation for total nitrogen content,ratio of total nitrogen to nicotine and physical properties among samples.The total nitrogen content decreased gradually along with the increase of single leaf weight,leaf length and leaf density;the ratio of total nitrogen to nicotine increased along with the increase of leaf length and leaf width,and decreased along with the increase of leaf density.Relationships of total nitrogen content could be described by inverse equation of（?）=0.827+38.047/x with leaf length,and compound equation of （?）=1.880 x 0.99^x with leaf density,respectively;relationships of ratio of total nitrogen to nicotine could be described by compound equation of（?）= 0.413×1.023^x with leaf width,and power equation of（?）= 2.791×x^-0.354 with leaf density,respectively.10.By using statistical analysis methods,124 samples of cutters in flue-cured tobacco leaves from main tobacco-growing areas of China were used as materials to analyze the relationship between the reducing sugar content（RS）,ratio of reducing sugar to nicotine （RS/Nic） and shape characters.The results indicated that there existed extensive variation among samples for shape characters and RS.The variation tendency was similar between RS, RS/Nic and single leaf weight that was generally decrease along with increase of single leaf weight,however it was increase after single leaf weight was beyond 12.3g.Similarly,the variation tendency was similar between RS,RS/Nic and leaf length that was increase along with increase of leaf length,especially,both rapidly increased after leaf length was beyond 64.9cm.The variation tendency was also similar between RS,RS/Nic and leaf thickness that was generally decrease along with increase of leaf thickness,and RS/Nic rapidly decreased after leaf thickness was beyond 0.126mm.Moreover,RS/Nic gradually increased along with increase of leaf width.The curve regression equations between RS,RS/Nic and each shape characters were built respectively.The relationship of RS could be respectively described by quadratic equation（?）=45.243-3.667x+0.139x²,（?）=11.217+316.779x-2026.595x² with single leaf weight,leaf thickness.Meantime,the relationship of RS/Nic could be respectively described by logarithmic equation（?）=16.278-3.375In x,compound equation （?）=14.075×0.002^x with single leaf weight,leaf thickness.All above equations were significant at the 5%level,and more efficient than simple linear regression analysis.The forecast effect was better after data validation.11.122 samples of cutters in flue-cured tobacco leaves from main tobacco-growing areas of China were used as materials to analyze the relationship between potassium content, chlorine content,ratio of potassium to chlorine and physical property.The results indicated that there existed extensive variation among samples of flue-cured tobacco leaves studied.The potassium content had respectively high significant positive correlation with leaf length, negative correlation with leaf density.The chlorine content had significant positive correlation with leaf density.The ratio of potassium to chlorine had respectively high significant positive correlation with leaf length,significant negative correlation with leaf thickness.The relationship of potassium content could be respectively described by inverse equation （?）=7.803-343.262/x with leaf length,by compound equation（?）=4.612×0.991^x with leaf density of cutters in flue cured-tobacco leaves from south tobacco-growing areas of China. Meantime,the relationship of chlorine content could be described by quadratic equation （?）=1.268-0.034x + 0.0003x² with leaf density.The relationship of ratio of potassium to chlorine could be respectively described by compound cquation（?）=0.023×1.105^x with leaf length,by quadratic equation（?）=94.408-1355.208x + 5897.907x² with leaf thickness.All above equations were significant at the 5%level,and more efficient than simple linear regression analysis.The potassium content and chlorine content decreased along with increase of single leaf weight, significantly increased along with increase of leaf length of cutters in flue-cured tobacco.The chlorine content increased along with increase of single leaf weight,rapidly decreased along with leaf length,increased along with increase of leaf thickness.These were two sensitive values of leaf width as 23.5cm,leaf density as 85.31 g/m² of cutters in flue cured-tobacco leaves from south tobacco-growing areas of China.12.In order to find the correlation between appearance quality and inner quality in flue-cured tobacco,the experiment studied the correlation between main chemical compositions and leaf length of different stalk positions in flue-cured tobacco by applying descriptive statistics,ANOVA,nonparametric tests,multiple comparisons and degree distribution.The results indicatcd that there existed extensive variation among chemical compositions and leaf length.The difference of chlorine,ratio of potassium versus chlorine among stalk positions was not significant in flue-cured tobacco.However,it is opposite of leaf length,nicotine,ratio of total nitrogen versus nicotine and ratio of rcductivc sugar versus nicotine.For total nitrogen and rcductivc sugar,the diffcrcncc between cutters and lower leaf was not significant,but it was significant between cutters,lower leaf and upper leaf.Similarly, for potassium,the difference between cutters and upper leaf was not significant,but it was significant between cutters,upper leaf and lower leaf.When leaf length was over 55.3cm,there was biggish change for nicotine,total nitrogen and ratio of total nitrogen versus nicotine of upper leave and lower leaf and for rcductive sugar of cutters and lower leaf.For ratio of rcductivc sugar versus nicotine of upper leaf and potassium,chlorine and ratio of potassium versus chlorine of cutters and lower leaf,there was also biggish change.13.K326,a flue-cured tobacco variety,was selected as materials and employed to study the correlation between main chemical components and physical properties of different leaf positions by using the method of canonical correlation analysis.The results indicated that the significant indexes of chemical components and physical properties were different with different leaf positions.The common significant indexes of different leaf positions including upper leaf,cutters and lower leaf were leaf length,leaf thickness and leaf density for physical properties,and chlorine content,potassium content and the ratio of potassium to chlorine for chemical components.The effects of common significant indexes of chemical components showed the same order of potassium＞ratio of potassium to chlorine＞chlorine among different leaf positions.The correlations of leaf length,leaf thickness and leaf density with potassium,chlorine and ratio of potassium to chlorine were significant for different leaf positions.The canonical loading coefficients could be as a reference for determining the significant indexes of physical properties and chemical components in flue-cured tobacco leaves.14.In order to advise the little blending leaf groups design of blending strips,the test founded GM（1,1） grey model and metabolic GM（1,1） grey model of ratio of total sugar to nicotine,and GM（1,1） grey mode,exponent smoothness GM（1,1） grey mode and logarithm GM（1,1） grey mode of ratio of total nitrogen to nicotine in aged flue-cured tobacco.The model test indicated that all model founded were significative and were applied in medium and long-term forecast.The model accuracy of ratio of total sugar to nicotine was higher and attained 98.37%and 99.73%,respectively.The model accuracy of ratio of total nitrogen to nicotine attained 93.56%and 88.52%after the GM（1,1） grey model was modified by exponent smoothness and logarithm.15.The test founded GM（1,1） grey model of total aroma contents,aroma quality and aroma quantity of aged leaf by grey theory.The model accuracy of total aroma contents,aroma quality and aroma quantity of Henan B2F attained 98.63%,99.24%and 99.91%,respectively. The model accuracy of total aroma contents,aroma quality and aroma quantity of Henan C3F attained 97.82%,99.94%and 99.09%.16.The most important problem of modeling was to correctly select the variable of model. To solve the problem,the test analyzed the grey model variable between main chemical components and aroma quality of flue-cured tobacco by integrated evaluation method both absolute degree of grey incidence and grey incidence.The results indicated that the variables of aroma quality modeling were total nitrogen,reducing sugar,ph values,the ratio of total sugar to nicotine and the ratio of total nitrogen to nicotine.The variables of aroma quantity modeling were total nitrogen,reducing sugar,potassium,ph values and the ratio of total sugar to nicotine The problem of relation between independent variable and dependent variable was solved by integrated evaluation method both absolute degree of grey incidence and grey incidence,so was the problem of multiple mutual linear of independent variable.17.The test founded GM（1,N）-BP neural network forecast model of aroma quality and aroma quantity.The results indicated that the model simulation and forecast accuracy of aroma quality attained 95.29%and 98.56%,respectively,and the forecast accuracy increased 3.4% than that of GM（1,N）.The model simulation and forecast accuracy of aroma quantity attained 95.13%and 98.20%,respectively,and the forecast accuracy increased 2.16%than that of GM （1,N）.When network structure was 1-3-1,the forecast accuracy of aroma quality was the highest.So was when network structure of aroma quantity was 1-9-1.18.In order to provide the base for computer-aided design blending,the tar of cigarettes from different batches of the same trademark as materials,were analyzed by time series analysis and the forecast model of tar of cigarettes smoke was built.The results indicated that the forecast precision of ARMA（2,2） model reached 99.51%and the average relative error was 0.49%,so the ARMA（2,2） model belonged to the first grade model（excellent model） and was superior to cause and effect model.Modeling by time series could escape the difficulty of independent variable choice of model,however,time series usually was used as short-term forecast and not long-term forecast.更多还原