Research On Soft-sensing Modeling Method Of Online Estimation Based On Vinyl Acetate Polymerization Rate

In order to ensure that production equipment is under the best operating conditions, and improve product quality and economic efficiency of enterprises in the industrial process control. It must be real-time detection and optimal control the important process variables that is closely related to product quality. The vinyl acetate polymerization rate is an important quality indicators in the process of poly vinyl alcohol production, it has an significant impact on economy, security of production process and products final usage. However, most of our factories can not be achieve continuous on-line detection of the vinyl acetate polymerization rate due to some technical or economic reasons. To solve this problem, an online prediction method of vinyl acetate polymerization rate using soft sensor technology based on least squares support vector machine (LSSVM) was advanced which is adopts the structural risk Minimization principle. The primary research works of the thesis are as follows:First of all, the basic principle of soft sensor technology and modeling methods are summarized, and obtained a variety of factors affect the vinyl acetate polymerization rate through in depth analysis of the poly vinyl alcohol in industrial processes. Then selection of auxiliary variables are researched deeply.Secondly, the basic principles of support vector machine, kernel function and least squares support vector machine regression algorithm derivation has been detailed studied, and establish the soft sensor model of vinyl acetate polymerization rate based on LSSVM. The simulation result indicated that the model based on LSSVM has more stronger generalization ability than that based on support vector machine and radial basis function neural network. However, the study found that LSSVM parameters has a great influence on the vinyl acetate polymerization rate soft sensor model performance, inappropriate model parameters will result in the vinyl acetate polymerization rate soft sensor failure.Finally, in order to resolve the disadvantages of traditional method select LSSVM model parameters, two quantum swarm intelligence optimization algorithms are proposed to select LSSVM model parameters automatically. The methods can converting the LSSVM model parameters of selection into optimization problem, the best parameters of LSSVM model would be selected by quantum genetic algorithm (QGA) and quantum particle swarm optimization (QPSO) that have the better global ability of search, and the best regularization parameter and kernel function parameters obtained by the optimization algorithm are used to establish vinyl acetate polymerization rate soft sensor model base on QGA-LSSVM and QPSO-LSSVM. Simulation results show that compared with the methods base on the traditional cross validation, genetic algorithms and standard Particle Swarm Optimization (PSO), the two models are proposed in this paper both have good prediction accuracy and generalization ability, well positioned to meet the industry control requirements of the vinyl acetate polymerization rate. However, compared with the QGA-LSSVM, the model based on QPSO-LSSVM is more suitable for soft sensor modeling of vinyl acetate polymerization rate Because of which can obtain the faster speed of modeling and smaller generalization of the absolute error.The results provides a new and effective means to achieve the online detection of the vinyl acetate polymerization rate.