Nonlinear Multi-Step Predictive Control Based On Neural Networks

Since the 1960's, the predictive controls have emerged as a powerful practical control technique especially in the process industry. Because of the adapt viability to the complex system, predictive control can conquer some kinds of problem in the industrial process, which would be a handicap of the modern control theory faced. With the development of microcomputer, the realization tools have been obtained; therefore, the predictive control has wide application and development.This thesis studies on the method of nonlinear multi-step predictive control based on neural networks. The generalized nonlinear predictive controller based on neural network is put forward by combining the theory of predictive control with generalized nonlinear PID method; Subsequently, the neural network inverse dynamic control method is introduced into predictive control. Then, the neural network based inverse control under the multi-step predictive index function is brought forward.The whole thesis include 5 chapters, and the main contents and conclusions are summarized as follows:In chapter 1, we review the history and development of the predictive control, some details of generalized predictive control and it's algorithm. And the stabilization and robustness of the system are also discussed. At last the predictive control based on neural networks are expounded.In chapter 2, we introduce the structure, theory and algorithm of the neural networks and also it's application in the predictive control field.In chapter 3, we put forward the method of nonlinear predictive PID control based on neural networks. In this method a neural network is construct to identify the nonlinear system and be as the recursive predictive model. A nonlinear PID controller is constructed by combining the predictive theory to control the nonlinear process. The simulation results prove the efficiency of this approach.In chapter 4, we present a method of neural network inverse dynamic control under the mutli-step predictive index function. The direct multi-step predictive approach and the recursive multi-step predictive method are adopted. The result is that the former is more exact than the latter and the simulation result prove the conclusion.In chapter 5, the conclusion of the thesis, our work and the future research are presented.