Research Of Mathematic Model Buiding For Cigarette Quality Control

The relationship, between cigarette's physical and chemical parameters and smoke's chemical parameters, has always been the focus of attention in tobacco science and technique research institutes. And it also is a significant factor of cigarette's quality control.The research target of this paper is cigarette's physical and chemical parameters and smoke's chemical parameters from a certain brand. In order to design a suitable mixed-level factorial experiment, a minimal-energy based genetic algorithm method is adopted and corresponding software is built.Model is abstracted on the base of research system. As for the model Y=F(x), a database technique is used to develop an analytical software, which integrates model management, sample management and modeling and prediction. This software can be used to select right subclass from the huge data to build model. In this software, five modeling measures such as MLR, PCR, PLS, Artificial Neural Network (ANN) and Genetic Algorithm-PLS (GA-PLS) are concerned. Genetic algorithm is used to select reasonable variable quickly as a huge number of variable concerned in modeling. Cross-validation is added into PCR, and PLS to check up odd sample point. A reliable model assessment means is also involved in this software.After experiment, according to the design, the modeling software can be applied to build model with experimental data. Thus some explicit relationships in cigarette parameter can be gained. Such results show the information that cigarette model is nonlinear. The average relative error results of nicotine, tar and CO operated by PLS, GA-PLS and BP algorithm are 7.5%, 6.3%, 5.9%; 5.8%, 4.4%, 5.2%; 5.5%, 4.8%, 5.7% respectively. When the model is used to predict the real experimental results, the average relative error results would be 9.1%, 6.1%, 4.7%; 5.1%, 3.4%, 4.7%; 4.4%, 3.1%, 4.2%. The verifying results illustrate that neural net prediction value is approximate to actual value. Then it can be concluded that the neural model is reliable, which reveals the variable relationship so as to direct cigarette products from the aspect of physical and chemical parameters. Thus the smoke's tar capacity can be controlled and decreased.