Study Of Nonlinear Predictive Control Approaches

With the development on theory and practice, predictive control has gradually formed the three methodological mechanism: model prediction, rolling optimization and feedback correction, and it is widely used in the industry. The study of nonlinear predictive control is also developing. This dissertation concentrates on the study of nonlinear predictive control approaches from the view point of predictive model and optimal method. The main contents are as follows:1) Neural Network Predictive Control. Based on the radial basis function network (RBF net) predictive model, a predictive control algorithm is proposed for nonlinear process. In this algorithm, orthogonal least squares learning method is used to establish the off-line process model, and the direct searching method is used to optimize the control variable. For single control horizon, we use the 0.618 method, and Powell's method is used for multiple control horizon. Through the simulation of a nonlinear process — pH-CSTR, the effect of single control horizon and multiple control horizon is compared. Then an adaptive algorithm is given. A recursive least square method containing the historical data is used for on-line identification of RBF net model. In the simulation, the process is set to the strongest nonlinear point, and the adaptive algorithm gets a satisfied result. And then a nonlinear predictive control based on the steady state error is proposed. In this algorithm, a RBF net is used as a steady state model. In order to improve the dynamic response, a modified algorithm is developed. Another RBF net used as a dynamic model is added to the system and a better effect is proved from the simulation. And then a nonlinear dynamic matrix control based on...