Hybrid Model Prediction And Optimization Control For Polymerization Process

Polyvinyl chloride(PVC)is an important resin material with complex reaction mechanism,large inertia and strong time-delay.In addition,the research on real-time monitoring and quality control of the reaction in the production process face some limitations.In the actual production process,conversion is the most important index that directly affect the product quality.Therefore,it is very significant to set up an accurate mathematical model for this batch chemical process.In this thesis,the actual production process of vinyl chloride polymerization is analysed systematically,and a serial hybrid prediction model combining mechanism model and data-driven model is proposed in order to solve the problem of the unknown polymerization rate.For the shortcomings of the traditional BP neural network,an conversion predictive model based on BP neural network optimized by genetic algorithm is presented.The real-time optimization based control strategy is proposed for the significant parameter index,namely the temperature of vinyl chloride polymerization.The main contents in this thesis are as follows:(1)The purpose and research status of PVC are analyzed,and several modeling methods are described.Furthermore,the polymerization process characteristics of vinyl chloride are analyzed and then the appropriate modeling methods are selected.(2)The mechanism and production technique of vinyl chloride polymerization are analyzed in depth and systematically.According to the conservation equation,reaction kinetics and thermodynamics model,the polymerization conversion mechanism model of vinyl chloride polymerization process is established.The relationship between the output and input of the mechanism model is simulated and analyzed to provide the basis for the subsequent hybrid modeling.(3)The serial hybrid modelling method is used to establish the hybrid prediction model of polymerization conversion in vinyl chloride polymerization.The hybrid model adopts the structure of mechanism plus data.The known part of polymerization process adopts mechanism model,including the conservation equation,dynamic model and thermodynamic equilibrium model.Due to the actual difficulty to measure the conversion rate,the data-driven neural network model is used to estimate it.(4)The weak points of the traditional BP neural network used in modeling the polymerization rate model are improved by using the global characteristics of genetic algorithm.The weight and threshold of the BP neural network are optimized by genetic algorithm to improve the model accuracy.(5)The polymerization temperature was an important parameter index in the polymerization process.So,based on the above hybrid model,a control strategy based on real-time optimization method was proposed for the polymerization temperature of vinyl chloride,which can effectively reduce the influence of uncertainty on the optimization and control results.Simulation results verifies the effectiveness of the proposed control method.