Predictive Control Based on Neural Networks of The Chemical Process

A predictive control algorithm based on neural networks is proposed for a class of highly nonlinear and uncertain systems. The predictive model is constructed by using feed-forward neural network in the algorithm, the Levenberg-Marquardt is incorporated into the back-propagation algorithm for training neural network off-line and modifying the predictive model online, The precision of model is raised and the control performance is improved significantly. The method is applied to control a CSTR reactor of nonlinear. The results illustrate effectiveness of the proposed approach.