Intelligent Predictive Control Theory Study Based On Neural Network

Predictive control algorithm was originally aimed at the linear system. It is effective to apply this algorithm to plants only with weak nonlinearity which can be viewed as a model mismatch and be overcome by virtue of robust design. However, as for plants with strong nonlinearity, if linear models adopted, there exists a non-negligible error between the predicted output and the actual output, which prevents implementing optimal control. Therefore, prediction and optimization must be based on nonlinear models.The research background being nonlinear systems widely existing in engineering, the research pivot being theoretical research and simulation, the dissertation studies a new method of DMC algorithm based on neural networks identification. The idea of this method is that the neural networks as the identified model of control plant produces predictive output, the control law is obtained by means of receding horizon optimized algorithm and thereby the predictive control of the nonlinear system is realized.During the course of the research, the following achievements are obtained: Firstly, the paper explores deeply the theory of dynamic matrix predictive control (DMC), analyzes detailedly its predictive model, its methods of revising feedback and receding horizon optimization, and its stability & robustness, studies how the correlative parameters influence on control effect, and finds out the problems of ordinary predictive control algorithms directly used in nonlinear systems. Secondly, for the characteristic of BP and RBF network, the improved methods for BP and RBF are proposed, the DMC algorithms based on the improved BP and RBF networks respectively are proposed. These two algorithms are applied to time delay systems. simulations show the validity and feasibility of the proposed new algorithms. Finally, this thesis applies predictive control to robot path planning, Gave has used the predictive control algorithm seeking control law the way track method. Simulation results shows that it performs well in path tracking and that the error is small.