Modeling And Optimization Of Melt Index Prediction For Propylene Polymerization Process Based On Grey System Theory

Polypropylene is playing a very important role in industry and our daily life, which makes the quality control of the product of the propylene polymerization process very crucial. As the most important quota in polypropylene production, the control and prediction of melt index is particularly important. In this article, the soft sensor prediction of melt index in polypropylene production is discussed. Aiming at the uncertainty of the process and incompleteness of the procedure information, grey system model (GM model) is employed to develop the soft sensor prediction model. And then the back propagation (BP) neural network and radial basis function (RBF) neural network are employed to modify the residual of the grey model. The methods of building models that are put forward in this article improve the structure of the sensor prediction model and improve the performance of prediction. These optimized models work quite well on the practical data from real industrial processes and provide several options for the melt index prediction application.Major works and contributions of this article are as following:(1) Aiming at the requirement of building model and scope of application of one dimension grey model and propylene polymerization process, GM(1,1) model is employed to build the soft sensor prediction model of melt index. Using the prior values of MI, the model predicts the following values. The result proves the suitability and efficiency of the GM(1,1) prediction model.(2) Considering the propylene polymerization process contains several factors, multi-dimension grey model is employed to build predicting model. An improved GMC(1,N) model is employed which performs better than the traditional GM(1,N) model in MI predict. The GMC(1,N) prediction model takes advantage of the inner relationship between the process factors adequately. The predict result based on practical data from real industrial process implied that the GMC(1,N) prediction model makes a better performance than that of SVM prediction model.(3) Based on the combination-modeling theory, BP neural network and RBF neural network are combined with GMC(1,N) model respectively. The residual modified models of both GMC(1,N)-BP model and GMC(1,N)-RBF model are built. The predict result based on practical data from real industrial process proves the suitability and efficiency of the combination models.