Predictive control based on neural networks and application to non-linear system

A predictive control algorithm based on neural networks is proposed for a class of highly nonlinear and uncertain systems. A feed forward neural network is used as the predictive. medel in the algorithm. The Levenberg-Marquardt isused for training neural network off-line and modifying the predictive model on-line. At each sampling time, the controller is optimized, so as to overcome the effect of model-mismatch and time-varying. The method is applied to control a CSTR reactor of non-linear The results illustrate the effectiveness of the proposed method.