Research On Nonlinear Model Predictive Control And Application

Model predictive control (MPC) is an example of successful applications of modern control theory in industrial processes. Research on linear predictive control has become mature and linear MPC has gained wide applications in industrial processes. However, most processes in industry are nonlinear, time-variant and bear uncertainty, thus research on nonlinear predictive control has become an important issue in control field. In industrial process control, for example, with the increasing demand of the quality and volume of products and the pursuit of economic profits, the processes are becoming more and more complex thus the control performance cannot improve without considering the nonlinear charicteristics in the control systems. Based on former research, the thesis gives a survey of MPC on the basis of related references and deals with some issues of nonlinear predictive control on a theroretical and practical background, the main contents contain the following four parts:1 Several predictive control design methods for a kind of nonlinear systems:(1) Focusing on the unsolved problems of neural network predictive control, the thesis presents a novel neural network predictive control method and extends the method to predictive functional control. This strategy uses only one neural network to design the nonlinear predictive controller, thus the control system is less complex and computing time is reduced. At the same time, the nonlinear multi-step predictions are converted into a series of linear predictions with an analytical control law.(2) A novel support vector machine (SVM) predictive control method is proposed focusing on the problems of SVM predictive control. The method gives a direct and effective multi-step predicting method by using the linearization of nonlinear kernel function and uses linear methods to get the control law which avoids the complicated nonlinear optimization.2 Predictive contol design for a class of bilinear systems.(3) A new SVM based multi-step adaptive predictive control algorithm for a class of bilinear systems is presented. The system is converted into a simple linear model by using nonlinear SVM dynamic approximation with an analytical control law. The method does not need on-line parameters estimation and the control method is a more accurate one compared with other traditional ones.3 Some predictive control design methods for a class of nonlinear mechatronic drive systems:(4) Extended state space predictive control method. The controller can control nonlinear nonminimum open-loop unstable systems with dead time. Its control performance is superior to predictive controllers based input-output models and state space predictive controllers with input-output cost function.(5) Adaptive predictive functional control (PFC) method. The method changes the nonlinear model into an equivalent time varying linear model. The structure of this method is similar to classic PI optimal controller. Its control performance and robustness are superior to traditional controllers for this kind of systems.(6) SVM based predictive control method. The method changes the nonlinear model into a global off-line linear model, thus on-line parameters estimation is not need. The structure of the controller is similar to classic PI optimal controller. Its control performance and robustness are superior to traditional controllers for this kind of systems.4 Application of PFC to industrial coking equipment.(7) A new kind of PFC-PID control method is proposed for nonlinear industrial coking equipment with applications to chamber pressure control of coking furnace.and liquid level control of fractionator. Real-time operations have shown improved control performance.The goal and control are separable in the control strategy and it effectively reduces the bad influence of nonlinear valve on the process performance and the stability, robustness and disturbance rejection of the control system are enhanced.