An Application Of Data Mining Technology In Polymerization Process

Data mining is an integrated technology based on data extraction. It is built upon database and does not rely on prior knowledge of objects so it can avoid error brought by prior knowledge. Data mining technology originates from commercial applications and is little used in industrial field, especially in the chemical field. This paper studies data mining and its application in the polymerization process based on data characteristics of chemical processes.Because the single model can not extract data information well due to the limited data fitting capacity of Support Vector Machines (SVM), the polymer quantity of polymerization process can not be estimated accurately. To solve this problem, classification tree combined with SVM model based on sample features is put forward so as to improve the estimation accuracy of polymer quantity. Classification based on sample data characteristics can avoid the diversification of the classification result. Simulations show that the compositional SVM model has higher estimation accuracy.Data classification brings more information interference between adjacent categories, which limits the model estimation accuracy and causes incapacity of tracking data with large hopping. The Linear Discriminant Analysis (LDA) is a method to expand the boundaries of classification. It can suppress the information interference between categories and improve sub-model's estimation ability. Simulations show that the model can not only suppress the information interference between categories but also improve the estimation accuracy with strong ability of tracking hopping data.The output data are true reflection to object characteristics and have great value for object changing. The existing data classification is generally divided according to input data while ignoring the output data. Bayesian classifier is based on both the input and output data and uses probability to choose data attributes. The LDA transformation based on the Bayesian classifier is to reduce the information interference between categories. The simulations show that the model has higher estimation accuracy and better tracking capability with practicality.The SVM gives all samples with the same punishment coefficient so some important samples cannot be highlighted. This paper finally proposes a compositional SVM model based on the AdaBoosting algorithm. On the basis of Bayesian analysis, this method initializes the penalty coefficient by using the Bayesian probability of samples. Then the penalty weight is updated by the use of loss function. So SVM training model can highlight some important samples to improve its estimation accuracy and generalization ability. Simulation shows that this approach can greatly improve the estimation capacity and generalization ability of the model.