Research On Model Identification And Nonlinear Predictive Control Of Continuous Chemical Processes

Model identification and control of chemical processes is a research hot spot in the field of process control. Chemical processes usually have time-varying and high nonlinear characteristics, so it is difficult to develop frist principle model of the processes. Control performance of temperature, product concentration and average molecular weights affect the quality and quantity of the products, so it is significant to research the chemical process model identification and control. Model predictive control (MPC) methods, which developed in practice, have been broadly applied in many fields. The traditional predictive control methods are usually based on the linear model. Satisfactory performances are difficult to be obtained using linear MPC algorithms because of the time-varying and high nonlinear characteristics of the chemical processes, so nonlinear model predictive control should be developed.Typical continuous chemical processes shuch as continuous strirred tank reactor, pH neutralization process and MMA polymerization reaction are used in the dissertation. Input-output data of the processes are used to identify the model to represent the system dynamic characteristics, and then the nonlinear model predictive controllers are developed based on the identified model. Also, particle swarm optimization algorithm is used to solve the predictive control law of the constrained MPC. Details of the contents are listed as follows:1. Step-responses and stability characteristics of continuous strirred tank reactor are analysised. Mechanism model of the CSTR is identified to a LSSVM-ARX Hammerstein model, and the identification results are compared with the traditional polynomial based Hammerstein model. A nonlinear model predictive controller is designed based on the LSSVM-ARX Hammerstein model, optimization of the nonlinear model predictive control is converted into an optimization problem of linear model predictive control. Predictive control law is deduced, and inverse model of the nonlinear part is constructed consequently. The proposed control scheme is used to control the product concentration of CSTR, simulation results are compared with the traditional nonlinear model predictive control and PI controller.2. A new method of Wiener model identification is proposed, Laguerre filters and least squares support vector machines (LSSVM) are introduced respectively as dynamic linear and static nonlinear blocks of the wiener model. Model structure and identification procedures are developed, and the method is extended to MIMO Wiener model identification. SISO and MIMO cases are performed, and the simulation results are compared with polynomial based Laguerre-Wiener models.3. Laguerre filters based Wiener model identification and nonlinear model predictive controls are applied to a pH neutralization process. First principle model of the process are identified to Laguerre-LSSVM Wiener models, Laguerre-SVR Wiener models, Laguerre-polynomial Wiener models and linear Laguerre models. Identification results are compared and the predictive control shemes are performed based on the four Wiener models, and SQP algorithm is used to solve the predictive control law.4. Gaussian process model identification and nonlinear model predictive control is applied to MMA polymerization process. Different model delays are analysed and the most proper one is chosen to develop Gaussian process model by comparing the predictive performances and computational loads. The identified Gaussian process model is used as the prediction model, and the nonlinear model predictive control algorithm is applied to the product average molecular weight control of MMA polymerization process.5. To avoid premature convergence of particle swarm optimization, a novel hybrid genetic particle swarm optimization algorithm is proposed. Eliminate threshold is set in the iteration of PSO, multi-offspring competition crossover and mutation operation are made between best particle and eliminated particle, the eliminated particle is substituted by the new particle which has better fitness value to improve the gloal searching ability of PSO. Based on the HGPSO, a constrained NMPC is proposed based on NARIMAX model. Online model parameters identification and predictive control law solution are developed and the NGPC algorithm is applied to an openloop unstable bioreactor.