Research On Model Identification And Nonlinear Predictive Control Methods Of CSTR Process

As a typical polymerization equipment of the chemical engineering process, the continuous stirring reactor (CSTR) has the characteristics of the strong nonlinear, the strong coupling, the uncertainty and the complicated dynamics mechanism, which make it difficult to establish an accurate mechanism model and affect directly the quality of the chemical products and productions, so the modeling and control of CSTR process has become a focus in the process control field. In recent years, the scholars at home and abroad have many systematic studies about the nonlinear predictive control based on the least squares support vector machine (LS-SVM), multivariable constrained, Hammerstein model, Wiener model, and T-S fuzzy model et al..These methods are applied to CSTR process or pH concentration process and many research results are achieved.Based on the CSTR process production control of as the background, the model identification and the design of the nonlinear predictive controllers are studied to solve some problems in the CSTR process about the key control parameters of the reactant concentration, the coolant flow, the pH concentration change, the reactor temperature affected by the reaction heat production, the reaction materials concentration, the discharge temperature including the characteristics of the strong nonlinear, the strong coupling, the uncertainty and the complicated dynamics mechanism. The main research works are summarized as follows:1. Based on Hammerstein-Wiener model including the input nonlinear module, ARX linear module and the output nonlinear module and the two nonlinear module modeled by LS-SVM with RBF, the identification method is proposed for solving the problems in the CSTR process about the reactant concentration and the coolant flow including the characteristics of the strong nonlinear, the strong coupling, the uncertainty and the complicated dynamics mechanism.Based on the identification model, combining with the predictive control principle, in the condition of the two nonlinear minimal performance index function in each control period, the nonlinear generalized predictive controls of the two nonlinear modules are obtained respectively in each control period by the trained predictive control input sequence using the neural network and by the solved nonlinear predictive control law using the quasi-Newton algorithm. The simulation results show that the controller has a good anti-interference ability and stability by the random sequence obtained from the CSTR process excitation inputs. 2. An iterative predictive control algorithm is proposed based on the interval control and multivariable constraints for the pH concentration change of CSTR process including the input and output strong coupling characteristics.At each sampling time the nonlinear model is transformed into the global linear model by mathematical transform method. The actual output is constrainted by the predict output interval.The predictive control principle is combined, and the minimum of the tracking error prediction is gained by the constrainted actual output and the revised tracking error forecast constraint in the objective function. The algorithm is applied to the pH concentration tracking process and the simulation results show that the controller has a good tracking accuracy and robustness.3. A constraint predictive control algorithm is proposed to solve the uncertain problem in the CSTR process about the reactor temperature affected by the reaction heat production including the state, the output, and the intermediate variable constraint. In the method, the state and the intermediate variable constraint is predicted by the constraint Hanmerstein model and the state observer in the nonlinear system and the interference of the nonlinear system is estimated by Wiener model and Kalman filter. The algorithm is applied to the reactor temperature affected by the reaction heat production process and the simulation results show that the controller has a good stability and robustness.4. A nonlinear adaptive robust predictive controller is proposed to solve the multivariable parameter uncertainty problem in the CSTR process about the control of the reaction materials concentration and the discharge temperature. The controller’s functions are to reduce the influence of the external disturbances by the estimated steady state error using a nonlinear disturbance observer (NDO), and to improve the stability of the closed-loop nonlinear system using a nonlinear adaptive predictive controller (NAPC), and then the target trajectory robust tracking is implemented. The algorithm is applied to the tracking and controlling of the reaction materials concentration and the discharge temperature processes under/without the condition of NDO.The simulation’s results show that the steady-state error of the closed-loop system can be quickly asymptotically converges to zero under the condition of NDO and improve the system’s tracking accuracy and stability.The paper supported by the National High Technology Research and Development Program of China (863Program)(No.2014AA041802) and the National Natural Science Foundation of China (No.61074020).