Nonlinear Predictive Control For Polypropylene Grade Transition Process Based On Neural Network Modeling

Polypropylene (PP) is a kind of synthetic resin polymerized by propylene and the key product of plastics industry. It possesses many great merits, such as good mechanical properties, corrosion stability, heat stability and easy to machining. Therefore, PP has been widely used in light industry, chemical industry, chemical fiber, building materials, electrical household appliance, packaging and motor areas. Nowadays it is one of the backbone industries of our country.Grade transition is the operation which translates one process condition to another. Considering on transition materials, huge economic loss is caused by frequent traditional grade transition which brought by various demands made by market and competition, therefore the optimization and control to the process of grade transition has great meanings. Taking Spheripol process mechanism as example, this thesis focuses on the modeling, optimization and control of grade transition process. The main research work and achievements are listed as follows:1. It dedicates the importance of research to the grade transition process by introducing the product condition of PP in and abroad and its supply and demand in current market. Then through technology characteristics of production, the knowledge about PP was systematically introduced.2. Taking Spheripol process mechanism as example, the static model of polypropylene process is established by modified nonlinear least squares and BP neural network respectively combined with production data, and the dynamic model is established by time-delay BP neural network. In all, the whole grade transition process is of BP-time-delay BP-BP structure. Simulation results show that the model can predict the melt index to a certain extent.3. A modified differential evolution (MDE) algorithm is proposed to optimize the trajectories of grade transition model, then neural network nonlinear predictive control based on MDE is employed to control the grade transition process on line. The proposed algorithm is applied in the polypropylene process, and the simulation results show that the performance of grade transition control can be greatly improved.4. Finally, the conclusions and forecasts are presented.