【作者】 胡建军；

【导师】 周冀衡； 张建平；

【作者基本信息】 湖南农业大学 ， 烟草科学与工程技术， 2009， 博士

【摘要】 烟叶原料是支撑中国烟草发展尤其是中式卷烟发展的基础,科学而准确地评价烟叶质量有助于指导烟叶原料的生产、采购和工业应用。然而,烟叶原料质量评价本身是一项复杂的系统工程,没有系统的烟叶质量评价方法体系,客观上将影响烟叶质量评价的科学性和有效性,导致烟叶质量评价的盲目性。因此,建立系统的烟叶质量评价方法,对准确地分析和认识烟叶质量的内在规律,简明扼要地提取烟叶质量的本质特征,清晰地展示烟叶质量的系统结构,科学地描绘烟叶质量系统的形成规律和变异特点,不仅具有理论意义,也具有现实意义。本研究综合了烟草原料学、烟草化学、烟草加工工程、化学计量学、系统工程、现代统计分析和数据挖掘等学科的理论与方法,形成寓于实践的烟叶质量评价方法体系,并以湖南和云南烤烟烟叶检测分析数据为样本,对所建立的烟叶质量评价方法进行了优化和实证研究。主要研究结果如下:(1)根据烟叶质量评价的阶梯层次性,从方法学的角度,创新性地建立了包括烟叶质量探索性、解析性、预测性和综合性评价的方法体系,阐明了各评价层次的基本思路、主要评价内容和具体评价方法,并对所建立的评价方法进行了优化和实证研究,优选出了分别在烟叶质量探索性、解析性、预测性和综合性评价中应用可取得良好效果的评价方法,为科学评价烟叶质量提供了方法与理论依据。(2)提出了一种新的灰色关联度分析方法,即基于品质工程原理的灰色关联信噪比分析法,综合平衡了关联系数的均匀性和平均值对灰色关联度的影响,并通过应用实例证明了新方法的有效性。(3)首次将有着强大统计和作图功能的R软件及其编程语言,全面应用于烟叶质量评价,并取得了较好的应用效果,为应用现代最新统计分析方法和数据挖掘技术对烟叶质量进行系统评价开辟了新的手段;(4)首次将稳健性统计量、箱线图、优势分析、分类与回归树、广义加法模型、多元自适应样条回归模型、两阶段聚类法、稳健性设计、鲁棒MM回归、鲁棒LTS回归、局布(Loess)回归等11种现代统计分析和数据挖掘方法全面应用于烟叶质量评价,优化和实证研究表明:①在烟叶质量探索性评价中,综合应用稳健性统计量和箱线图,既能够准确描述烟叶质量的主体信息特征,又不会忽略对非主体信息的充分挖掘;②在烟叶质量解析性评价中,应用优势分析能够准确地识别烟叶各理化特性对烟叶质量的线性相对重要性,应用分类与回归树分析能够识别基于线性和非线性关系的烟叶理化特性的重要性,通过广义加法模型的偏拟合图可以清晰地展示烟叶质量与各理化特性间的线性和非线性关系;③在烟叶质量预测性评价中,多元自适应样条回归模型取得了较好的预测效果。作为分段线性回归,模型依赖的假设较少,既便于对模型的解释,又可以自动筛选关键变量,具有较好的应用前景;④在烟叶质量综合性评价中,两阶段聚类法适用于任何类型的变量,能自动确定最佳的类别数,并可以自动给出各评价指标在聚类中的重要性,有利于明确各聚类的重要性状、变异特点和类型间的分类特性,具有较好的应用前景;⑤应用应用稳健性设计对加工工序质量进行综合评价,能准确识别影响工序加工质量的重要因素、调节因素、稳健因素,次要因素,有利于提高并稳定加工质量。(5)全面系统地解析了烤烟质量与各类理化特性间的整体相互关系,结果表明:烟叶主要物理特性对外观质量的影响大于主要化学成分的影响;烟叶主要化学成分对评吸质量和主要烟气成分的影响大于主要物理特性的影响,而烟叶外观质量对评吸质量和主要烟气成分的影响最小;影响烟叶外观质量和评吸质量的因素远比影响主要烟气成分的因素复杂:烟叶的理化特性对烟叶质量的影响程度存在明显的地域性,不同产地的烟叶特色具有不可模仿性。(6)全面系统地解析了湖南烤烟单一质量与各理化特性间的线性和非线性关系,结果表明:可用于表征烟叶质量的关键理化指标有:总氮、还原糖、烟碱、有机钾、氮碱比、糖碱比、钾氯比、单叶重、叶片厚度、叶宽、叶片结构、身份、光滑与微青程度等;烟叶大部分香味成分与评吸质量之间存在显著的线性关系,也有部分香味成分与评吸质量之间存在明显的非线性关系;烟叶大部分香味成分与评吸质量之间在明显的正相关关系,但也有部分香味成分与评吸质量之间存在明显的负相关关系。(7)建立了拟合和预测效果较好的烤烟烟气焦油量预测性评价模型和烤烟香气质量预测性评价模型,以及综合评价效果较好的烤烟烟叶化学品质综合评价模型、烤烟品种可用性灰色综合评价模型和基于两阶段聚类法烤烟外观质量综合评价模型。更多还原

【Abstract】 High-quality tobacco-leaf raw-materials is the foundation to support the development of China’s tobacco, in particular for the development of Chinese style cigarette, and the accurate evaluation of tobacco-leaf quality will help to guide production of tobacco-leaf raw-materials, procurement and industrial applications. However, the tobacco-leaf quality evaluation is a complicated systematic project, there isn’t systematic evaluation methodology of tobacco-leaf quality and which will impact on evaluating the tobacco-leaf quality scientifically and effectively, leading to blindness of tobacco-leaf quality evaluation. Therefore, the establishment of tobacco-leaf quality evaluation method system not only has theoretical significance, as well as has practical significance, for accurate analysis and understanding of the inherent law of tobacco-leaf quality, extracting succinctly the essence characteristics of tobacco-leaf quality, showing clearly the structure of tobacco-leaf quality system, and depicting scientifically the formation laws and variation characteristics of these tobacco-leaf quality system. In this study, a combination of the theory and methods of lots of disciplines such as tobacco cultivation, tobacco chemicals, tobacco processing, chemometrics, systems engineering, modern statistical analysis, data mining, and so on, constitute the methods system of tobacco-leaf quality assessment based on practice. The methods of tobacco-leaf quality assessment were optimized and proved by the flue-cured tobacco samples in Hunan and Yunnan. The main results are as follows:(1) According to the ladder-level of tobacco quality evaluation, from the methodological point of view, the tobacco-leaf quality evaluation methods system was established innovatively including exploratory, analytical, forecasting and comprehensive evaluation methods system. The basic ideas, main evaluation content and main evaluation methods of the evaluation levels were clarified. The established evaluation methods were optimized by empirical research, and the optimization evaluation methods can be applied in the exploration in the quality of tobacco leaves, analytical, forecasting and comprehensive evaluation of the exploratory, analytical, forecasting and comprehensive evaluation of tobacco-leaf quality, which can obtain good evaluation results. The tobacco-leaf quality evaluation methods system provides methods and theoretical basis for scientific evaluation of tobacco-leaf quality.(2) A new gray relational analysis method, that is, the grey relation signal-to-noise ratio method was put forward creatively, based on quality engineering principles. The impact on the gray relational degree from the uniformity and average of gray relational coefficient, is balanced in overall by the new method. of correlation of gray, and through application of proven The new method is effective. The effectiveness of the method was proved by the application of examples.(3) The R software and its programming language, which has a good strong statistical and mapping functions, was used comprehensively in the tobacco-leaf quality evaluation, and the good application effect was obtained. It has opened up a new means to evaluate systematically the tobacco-leaf quality by the latest modern statistics analytical methods and data mining technology.(4) The modern statistical methods and data mining technologies such as robust summaries, box-plot, dominance analysis, classification and regression trees, generalized additive model, multiple adaptive regression splines, two-step cluster analysis, robust design, robust MM linear regression, robust LTS linear regression, local (Loess) regression, and so on, were introduced to the tobacco-leaf quality assessment on the first time. The main results of cases study are as follows:①We can accurately describe the main characteristics of tobacco-leaf quality information, and will not ignore to fully mining the non-main information of tobacco-leaf quality, if we make comprehensive use of the robust summaries and box-plot in the exploratory evaluation of tobacco-leaf quality;②In the analytical evaluation of tobacco-leaf quality, we can accurately identify the relative importance based on linear about the physical and chemical characteristics effecting on the tobacco-leaf quality by dominance analysis. The relative importance based on both linear and nonlinear relationship about the physical and chemical characteristics can be accurately identified by classification and regression trees. The linear and nonlinear relationship between tobacco quality and physical or chemical characteristics can be clearly showed through the partial fitted plot of generalized additive model;③A better prediction effect was obtained in the forecasting evaluation of tobacco-leaf quality by multiple adaptive regression splines. This is a piecewise linear regression model based on less assumptions and is easily explained. It also can identify the key variable. So its application has good prospects in the forecasting evaluation of tobacco-leaf quality;④In the comprehensive evaluation of tobacco-leaf quality, by two-stage clustering method applicable to any type of variable, the optimal number of clustering can be automatically determined, and the importance of the evaluation indexes in clustering can be automatically given, which will help to confirm the important characteristics and variation of any classification types. So its application has good prospects in the comprehensive evaluation of tobacco-leaf quality;⑤The processing quality for tobacco leaves was evaluated comprehensively by robust design and the factors impacting on the quality of processing can be accurately identified as important factors, regulating factors, stable factors or secondary factors, which will help to improve and stabilize the quality of processing.(5) The all and the one relationship between physical and chemical characteristics and the flue-cured tobacco-leaf quality was parsed comprehensively and systematically. The main physical characteristics’ effect on the appearance quality of flue-cured tobacco-leaf is greater than the impact of its main chemical compositions; The main chemical components’ effect on the main smoke gas components and the smoking quality of flue-cured tobacco-leaf are greater than the impact of its main physical properties, and the effect of the appearance quality is the smallest; The effect factors of smoking quality and appearance quality of flue-cured tobacco-leaf are more complex than the ones of the main smoke gas composition of flue-cured tobacco-leaf; There is significant difference for the degree of physical and chemical properties impacting on tobacco-leaf quality in the different planting tobacco regions and the characteristics of tobacco-leaf quality can not be imitative.(6) The linear and nonlinear relationship between the single quality and physical and chemical characteristics of flue-cured tobacco-leaf in Hunan was parsed comprehensively and systematically. The results show that: the quality of tobacco leaves can be tokened by these key physical and chemical indexes such as total nitrogen, reducing sugar, nicotine, organic potassium, ratio of total nitrogen to nicotine, ratio of total sugar to nicotine, ratio of potassium to chlorin, single-leaf weight, leaf thickness, leaf width, leaf structure, body, degree of leaf smooth and tiny-green, etc.. There is significant the linear relationship between the majority of flavor components of tobacco and smoking quality of tobacco, but there is obvious non-linear relationship between some flavor components of tobacco and smoking quality of tobacco. There is clear positive correlation between the majority of flavor components of tobacco and smoking quality of tobacco, but there is significantly negative correlation between some flavor components of tobacco and smoking quality of tobacco.(7) The flue-cured tobacco smoke tar forecast model and the flue-cured tobacco aroma quality forecast model were established, and the good fitting and prediction effects were obtained by these forecast models. The comprehensive evaluation model of flue-cured tobacco chemical quality, the gray comprehensive evaluation model of the availability of flue-cured tobacco varieties, and the two-stage clustering evaluation model of the appearance quality of flue-cured tobacco, were established, and the good comprehensive evaluation of the effectts were obtained by these comprehensive evaluation models.更多还原