Study On Predictive Control Algorithms And Their Applications

The history and state of Model Prediclive Control is summarized. The principle of Predictive Control is introduced brielly. In view of the importance of solving Diophantine equation for Predictive Control, the details of its computer-solving algorithm is discussed and the flow chart is given correspondingly. On the basis, the following contributions are made in this paper: 1. Based on the satisfactory control (compared with optimal control), the relationships between Goal Programming and Predictive Control is built aiming at complicated industrial process control. The three problems about Predictive Control with wide significance are solved : ?online receding- horizon optimization of Predictive Control with constraints, multi-goals and multi-freedoms; ?controllability of Predictive Control; ?online identification of parameterized model. 2. Since plant model based on impulse response can be easily obtained, it is used widely in industrial process control. Constrained increment Model Algorithmic Control strategy with dead time is studied based on the impulse response of plant. 3. In the practice of industrial process control, most plant models are hardly solved by analysis method because they have multi-goals, multi-variables and constraints. So how to find a fast and effective solving method to solve them is quite a significant research task. On the basis of plant step response, constrained Dynamic Matrix Control with multi-variables is studied. 4. Because of least parameters in the model, Generalized Predictive Control has light computational burden in online receding horizon optimization. Aiming for discrete transfer fi.inction, on the basis of salving Diophantine equation, constrained Generalized Predictive Control is studied by using Quadratic Programming and Goal Programming. Compared with Quadratic Programming, Goal Programming has many advantages such as fast computational speed and good numerical performance because it is a linear op~mal strategy. Ix 5. A Model Predictive Control algorithm with superfluous parameters is presented by incorporating the non-parametric model which is easy to be obtained and Generalized Predictive Control with less computational burden. This algorithm significantly reduces on-line computational burden comparing with non-parametric model. Because of the superfluous parameters in the model, it is unnecessary to get the exact priori knowledge concerning the stmcture and dead time of system. This property is not only convenient for modeling, but also useful for enhancing the accuracy and robustness of predictive model. 6. A Predictive Control algorithm is presented by incorporation of online identification and offline identification. Then aiming for the heat-exchanger outlet temperature plant of city hot-water networks and melting salt temperature plant of fixed bed reactor, the computer simulation study is made.