Rule Extraction And Formula Maintenance Of Cigarette Based On Bayesian Network

The maintenance of cigarette formula directly affects the quality of cigarette products and determines the characteristics of the cigarette brands,which plays an important role in the cigarette products manufacturing process.One or more kind of unblended cigarette often need to be replaced because of the shortage of unblended cigarette or the excessive cost in the process of maintaining cigarette formula.So far,the development and product maintenance of the tobacco product mainly rely on the human sensory evaluation of the experts,which means that whether the sensory indicators of the cigarette fulfill the requirements or not is mostly determined by the score given by the experts.When a certain unblended cigarette is missing,an expert will firstly use personal experience to select one kind of unblended cigarette to take the place of the missed one,and then compare the replacement with the previous one after smoking evaluation.After adding the replacement to the incomplete formula,the expert will evaluate its sensory indicator and observe whether it is similar to the original one or not.This kind of method is not only susceptible to subjective feeling from the expert,but also it is unable to meet the fast growing of production needs.Therefore,how to automatically recommend the similar unblended cigarette for the missed one and keep the basic stability of sensory quality and smoke indices is a very important part of the cigarette maintenance process to achieve automation.This thesis firstly extracts the rules after establishing a cigarette formula recognition model based on the historical data of cigarette formula by using the Bayesian network.And then a heuristic rule-based one to one replacement maintenance method of cigarette formula is proposed on the basis of combining the previous established formula recognition model with the clustering algorithm.Finally,the unblended cigarette replacement process of formula maintenance is integrated into an computer-aided decision-support system of cigarette optimization by using Matlab GUI.The main work of this thesis is given as follows:(1)Firstly,the supplement and discretization of the historical data of cigarettes formula are conducted based on their own characteristics.Then the correlation analysis and normal analysis of the chemical elements are carried out by using SPSS,Minitab and other statistical software.Finally,the statistical analysis of the frequency,the fluctuation of the position,the color,the quality level and the change of the physical and chemical indicator of unblended cigarette in the formula are conducted.(2)Firstly,based on the integrated data of cigarette formula,the recognition model of cigarette formula is established by using Bayesian network.Combining with correlation analysis about chemical elements,the result is compared with the one of the Naive Bayes model and the decision tree model,and then the advantages and disadvantages of each method in rule extraction are analyzed.Finally,Bayesian network is selected to extract the rules of cigarette formula which is important in the cigarette blending formula design.(3)Firstly,the weighted clustering method is conducted to analyze all unblended cigarette used in the brands and the amount of the unblended cigarette in the brands is counted out.Secondly,a heuristic method of cigarette formula maintenance is proposed on the basis of the Bayesian network cigarette formula recognition model,which achieves the goal of the automatic recommend of unblended cigarette one to one.After that,the best dosage is selected as the final amount of replacement after the sensory evaluation on the previous and later substitution.Then the replacement of tobacco is prioritized according to the probability obtained by the parameter learning in the Bayesian network and the similarity of sensory quality.Finally,the unblended cigarette replacement process of formula maintenance is integrated into an computer-aided decision-support system of cigarette optimization based on the Matlab GUI software environment.