【作者】 赵亮；

【导师】 雒兴刚；

【作者基本信息】 东北大学 ， 系统工程， 2017， 硕士

【摘要】 随着经济全球化的冲击,卷烟行业间的竞争日益激烈,对于卷烟产品的维护与开发要求越来越高。在我国,卷烟配方生产过程中需要大量的人工经验与配方实验,因此对卷烟工作人员的技术积累和经验要求很高,同时大量生产试验会造成过多的原料浪费。显然,这种方法效率低下、稳定性较差。基于卷烟企业生产过程中已经积累了大量历史数据,企业迫切需要对卷烟配方维护工作进行优化,探索出一条节约、简化、高效的维护方案。当前卷烟企业在卷烟配方维护过程中,仅依靠专家对卷烟烟叶进行人工评吸,在评吸的过程中不断调节各种烟叶的投放量。但是,这种方法的问题是工作量巨大且重复性较高,大量损耗时间及烟叶原料。企业需要一个能够结合历史数据进行自动推荐的配方维护系统。本文基于H集团提供的配方数据,进行了配方属性分析、配方规则挖掘与维护等相关研究,主要研究工作包含以下几个方面:(1)针对H集团提供的配方数据及其单料烟信息,对数据进行了预处理,补齐了数据中的缺失值,以SPSS软件为工具,对卷烟配方的化学元素、属性特征以及之间的回归关系进行比较分析,实验证明配方内部存在一定的相互作用关系。(2)基于配方知识的研究,发现配方当中存在大量抽象专家经验,利用关联规则方法进行挖掘,得到了配方中隐含的一些单料烟配伍规则以及单料烟共存关系。由于规则数量较多,进行了规则约简以便于使用。(3)在关联规则挖掘的基础上,建立了基于关联规则频繁项集和相似度方法的启发式卷烟配方维护方法,以具体的算法表示专家经验,动态地调配单料烟的配伍与投放量,实现卷烟配方维护的智能化。(4)在上述工作的基础上,结合企业实践需求,设计并开发了基于Matlab GUI的卷烟配方优化辅助决策系统,通过建立三个不同模块,实现了成品烟配方规则输出、卷烟配方启发式维护、成品烟配方指标辅助预测等主要功能,并且使用H集团数据进行了实例验证。更多还原

【Abstract】 With the impact of economic globalization,the competition of tobacco industry is increasingly fierce.The demand of maintenance and development of the cigarette products is higher and higher.In our country,a lot of expert experience and formula experiment are needed in the production process of cigarette formulation,so the cigarette staffs need strict technical accumulation and experience,at the same time a large number of production tests will cause excessive waste of raw materials.Obviously,the efficiency of the traditional method is low and has poor stability.Based on a large amount of historical data accumulated in the process of enterprise production,enterprises urgently need to optimize the formula maintenance,and to find a saving,simplified and efficient solution for maintenance.In the process of the cigarette formulation maintenance,cigarette enterprises presently only rely on experts to carry on the manual evaluation of the tobacco leaf..In the process of constantly smoking,they wil adjust the proportion of various kinds of tobacco leaves.However,the problem of this method is that the workload is huge,repetive,time-consuming and tobacco-wasting.Enterprises need a system that can automatically recommend the formulation based on historical data.Based on the formula data provided by H group,this thesis carries out the research on the correlation between the formula attribute analysis,the rule mining and formula maintenance.The major work includes the following four parts:(1)According to the data of formula and single cigarette provided by the H company,this thesis carries on the data preprocessing and fills the missing values in the data.By using SPSS software,the thesis explores the characteristics of cigarette formula between chemical elements and attributes,then compares the relationship of regression.Experiments support the internal relationship in the formula.(2)Based on the study of formula knowledge,a lot of abstract expert experience can be found.By using the method of association rules,some implicit compatibility rules and single cigarette coexistence relationships can be extracted.Because the number of rules is big,the rules are reduced for facilitating use.(3)Based on the association rule mining,the heuristic cigarette formula maintenance method is developed based on frequent itemsets and similarity method.It can be used to express the expert experience,dynamically allocate the quantity of the unblended cigarettes,and realize the intelligence of the cigarette formula maintenance.(4)On the basis of the above work and combining with the enterprise needs,an computer-aided decision-support system of cigarette optimization using Matlab GUI is developed.Three modules are built in the system.The first module is mining rules about formula,the second is heuristic maintenance and the last is formula indicators prediction.Dataset from H group were used to verify the deveoped algorithms.更多还原