Predictive Control Method On PVC Stripping Process Based On SVM

Poly vinyl chloride (PVC) industry occupies an important place in the national economy, and vinyl chloride monomer (VCM) is the main material to produce PVC. After VCM polymerization, it is necessary to strip PVC to remove the unwanted VCM for protecting the environment and reducing production cost. However, just as most of the plants systems in the industry field, PVC stripping process is nonlinear and simple control method will not perform perfectly. Support Vector Machine is a new learning method in the field of statistics. It is a powerful tool in nonlinear system identification and provides a new control theory which is suitable for complex nonlinear system. Generalized predictive control is one part of predictive control which has been widely used in industrial process control areas.In this paper, some researches have been done on the base of several field researches and reading many references about support vector machine and predictive control.Firstly, analyze and summary the PVC stripping process according to the actual JinHua Group conditions, and then find the problems and propose solutions from theory to practice.Secondly, the key to study is support vector machine regression system modeling and how to realize by MATLAB according to the SVM theory. Then build the PVC process model based on SVM, and the field data prove the validity of the model.Finally, a generalized predictive control implicit algorithm based on support vector machine model for a family of complex systems with strong nonlinearity is presented. In the process of system operation, the model is linearized at each sampling insant, and the GPC implicit algorithm is employed to implement the controlled plant. The simulation results show the effectiveness of the presented algorithm and analyze the main parameters in the algorithm. Then, after combinating of the online corrected model and error feedback compensation, simulation based on field data for PVC stripping process was conducted according to different actual conditions, and the results prove the validity of the algorithm.