Application Of Data Mining Technology In Optimization Of Cigarette Products Formula

The optimization design of formula product is a complex comprehensive technology, involves many subject fields. In order to obtain the formula which has superior performance and can satisfy the use requirements, according to the performance requirement of products and the technological conditions, determines each raw material the amount used allocated proportion relations, through the experiment, the optimization and the identification, selects raw material reasonably. For such a complex multi-objective formula system, the design of the test method is especially important. In recent years, the Applied Research of the formula optimum design is very active. Faced to so many design methods, how to select reasonable method had become a difficult problem to formula designers.To the features of the data gathered from cigarette industry, this thesis studies on the application of data mining technology on the optimization of cigarette products formula. It takes use of classical multi-variant statistical analysis methods to analyze the industry data and feature extraction, which is used to analyze the relation among the variants form different angles and planes. The model which is built with M5'model tree is in the form of expression, which could be accepted easily by the common users. And this model can be applied in the tobacco sensory evaluation. Support vector machine (SVM) have great ability of nonlinear approach, which are used to set up learning model for forecasting the sensory quality. SVM is applicable to deal with high dimension and small sample problem. Finally, the thesis takes advantage of genetic algorithms, based on the relevance of tobacco leaf physical and chemical composition with cigarette sensory and flue gas index, identify constraints design optimization model to achieve the goal of the lowest cost. By operating the optimized model, can obtain the required cigarettes style with best ratio combinations. At the same time, it can estimate the designed cigarette blending formulation ahead of time. To realize the key control to the key target, and make the design more science and rationality.