Fast Algorithm for Adaptive Multivariable Generalized Predictive Control

Abstract

A very concise method is presented to simplify the implementation of adaptive multivariable generalized predictive control (MGPC). If a physically realizable multivariable process can be described by a controlled auto-regressive integrated moving average (CARIMA) model with diagonal matrices C and A, the way to get MGPC controller coefficients can be simplified, since there exists direct expressions describing the nonlinear relationship between the open-loop model parameters and the MGPC controller coefficients according to a certain set of tuning parameters. The control moves are just the product of the process known information and the MGPC controller coefficients. Then a multilayer feedforward neural network is trained to obtain the controller coefficients from model parameters quickly, which substantially abates the mathematical computational overhead associated with MGPC. The feasibility and efficiency of this algorithm is demonstrated by comparison experiment results