Multivariable Nonlinear Process Control

The problems on the multivariable nonlinear process control are studied in this paper. The main works and research results consist of five parts: 1. After introducing neural network based nonlinear predictive control for processes with time-delay, the paper presents several on-line feedback correction methods. Then the neural network based multiple feedback predictive control, a variable correction coefficient and its design approaches are proposed for nonlinear cascade industrial processes. Further, this method is extended to multivariable nonlinear systems. The neural network based multivariable nonlinear predictive control with the intermediate state feedback and the design method of feedback correction matrix is presented. 2. The concept on static inverse model of dynamic multivariable nonlinear system and inverse system based linearization method of nonlinear system based on an integrating model is proposed. Several effective and practical ways obtaining static inverse model are presented. 3. The SISO Hammerstein model is extended to MIMO system. The generalized Hammerstein model and its identification method is proposed. Further, for general multivariable nonlinear system, an integrating model combining ANN with linear discrete differential model and its identification training method is presented. This kind of models based multivariable nonlinear predictive control algorithms are proposed. The algorithms employ the results of the linear predictive control and don't need on-line numerical optimizing which is necessary in general nonlinear model (including ANN model) predictive control. That greatly decreases on-line computing consumption, strengthens the reliability of the algorithm and the stability of the system. They are really effective control algorithms for the multivariable nonlinear system. 4. The design method of a compensator which can make an object optimally decoupled via the prediction of system outputs is proposed, and the algorithm of the compensator is presented. The predictive optimal decoupling compensator is different from traditional dynamic or static decoupling compensation, and suitable for most of multivariable systems including some particular ones. It has good stability and is easy to be realized. Further, the principle of the linear model...