Neural network model based predictive control for multivariable nonlinear systems

A nonlinear model predictive control (NMPC) algorithm based on a BP-ARX combination model is proposed for multivariable nonlinear systems whose static nonlinearity between inputs and outputs could be obtained. The dynamic behavior of the system is described by a parameter varying ARX model, whose parameters are estimated on-line with recursive least-squares algorithm and rescaled properly according to a BP neural network representing the system static nonlinearity. The construction of the BP-ARX model and a constrained NMPC algorithm based on the BP-ARX model are elaborated. The effectiveness of the proposed method is demonstrated by simulation on a multivariable chemical reactor system.