Polymerization Process Of Hybrid Modeling Method

Process modeling is of great significance in chemical industry process,and it is the important foundation of the process control and processoptimization. The polymerization process is characterized by complicatedreaction mechanism, and a nonlinear time-varying. These increase thedifficulty of process modeling. Aiming at the complex and continuousreaction process, the paper explores the hybrid modeling method, includes thefollowing contents:1. The paper uses styrene polymerization reaction process as background,discussing a hybrid modeling method based on radial basis function neuralnetwork. The reaction mechanism of polymerization process in this paper isknown, in which the reaction rate equation is unknown; the response variablecan be measured, including the reactant and product concentrations; reactionparameters which exist in the reaction rate equation are unknown. In view ofthe above characteristics, this paper uses approximation properties of theradial basis function. So the paper uses the radial basis function neuralnetworks to approximate the unknown part of the reaction rate equation model, and uses white box model to establish known partial differential equationmodel, and finally two models are connected into a whole to form grey boxmodel.2.The response variables cannot be measured on line for a class ofreaction, this paper presents a method based on minimum error methodto estimate the optimal value of the unmeasured variable on line. Makingfull use of the known condition, reasonable hypothesis, according tominimum error, the paper gets he numerical variables of the unmeasuredvariable on line.3. Using polypropylene reaction process as the background. Thecharacteristics of the polypropylene reaction process in this paper are asfollows. Reaction mechanism are partially known, wherein the reaction rateequation is unknown; The response variable portion can be measured, inwhich the monomer and the chain of free radical concentration can bemeasured; Reaction parameters which exist in the reaction rate equation areunknown. Firstly, the paper uses the minimum error method to estimation theoptimal value of the monomer and the chain of free radical concentration, andthen according to the estimated optimal values, the paper gets reaction ratedata samples which are used in radial basis function neural network. Secondly,according to reaction data of the output of the neural network and the knownreaction mechanism, the paper constructs the differential equation model.Polypropylene reaction process is more closed to the actual reaction process, The hybrid modeling method explored in the paper solves themodeling problem for the continuous polymerization process which containsunmeasured variables on line. At the end of the article, It points out that thehybrid modeling method used to be studied--model convergence problem.