Reduced Neural Model Predictive Control Strategies for a Class of Chemical Reactors

抄録

A simple model predictive control strategy based on reduced feedforward neural network (FNN) models is proposed. Under some physical constraint conditions, the short-prediction-horizon predictive control algorithm can carry out the offset-free performance for a class of nonlinear systems with input/output multiplicities. The main issue is to specify the input/output patterns for neural network architecture, and a stable, minimum-phase mode is added to reduce control structures that involve off-line identification algorithms and graphic-based determination. Finally, three examples of chemical reactors exhibiting unstable or nonminimum-phase dynamic behaviors are demonstrated to verify the proposed control scheme.