Bionics Intelligence Model Based On RBF Neural Network And Its Application In Propylene Polymerization Process

Melt Index (MI) is one of the most important parameters in determining the polypropylene's grades as well as the qualities of the different grades. Since the lack of the proper on-line instruments, and the difficulties during the using of these on-line instruments, it can be only gained by artificial sampling or offline testing, which makes the MI's measurement interval and delay are both very long and results in the difficulty of the quality control. MI prediction therefore becomes one of the frontiers and focuses in olefin polymerization research field.Considering the nonlinearity and the correlation of the input variables in Propylene Polymerization Process, the key of this thesis is to build up optimal MI prediction model with combining Statistic Modeling methods and Bionics Intelligence approaches to eliminate the modeling mistakes caused by parameter selecting and other artificial factors.The main work and contributions are listed as follows:1. Propylene polymerization process, polypropylene, especially the MI and its research development are introduced and reviewed. The knowledge of Statistic Modeling and Bionics Intelligence are further introduced.2. Considering the nonlinearity and the correlation of the input variables in Propylene Polymerization Processes, PCA-RBF model based on Radial Basis Function (RBF) neural networks and Principle Component Analysis(PCA) is built up to infer MI of polypropylene. The results of the research based on data from real Propylene Polymerization Process indicate that the explored PCA-RBF model can not satisfy industrial production standard.3. Chaos Optimization Algorithms based on Logistic Function is introduced to PCA-RBF model to optimize the key parameters, and PCR (PCA-CHAOS-RBF) model is proposed to infer MI of polypropylene. The research results based on data from real Propylene Polymerization Process show that the proposed method provides promising prediction reliability and accuracy.4. By reviewing the development, principles, characteristics of Genetic Algorithms(GA), with its good global optimization ability, PCGR (PCA-CHAOS-GA-RBF) model is therefore presented. The research results based on real process data prove the excellent prediction and generalization ability of the proposed PCGR model. In addition, PGR (PCA-GA-RBF) model, where GA is employed to optimize PCA-RBF model, is further presented and compared with PCGR model. The research results based on real process data prove the similar prediction reliability and accuracy between PCGR model and PGR model, which indicates the good global optimization ability of GA in MI prediction processes.5. The detailed comparative study among the above proposed different methods and the recent development in MI prediciton of PP research in the open literatures at home and abroad confirms the excellent prediction accuracy and generalization ability of the above proposed PCGR model and PGR model. Compared with the best reported RMSE result, 1.51%, reported by Professor Han in Journal of Applied Polymer Science, an increase of approximately 21.2% in generalization accuracy is obtained when the proposed PCGR model is applied, while an increase of approximately 17.2% in generalization accuracy is obtained when the proposed PGR model is applied.