Constrained multivariable model algorithmic control and its non minimum phase characteristics

Abstract

During the last two decades, Model Predictive Control (MPC) techniques, all having the common ida: the control input signal is based on a prediction of their process output, have been analyzed and applied in both academic and industrial environments. One such technique, Model Algorithmic Control (MAC), culminating in the recent marketing of a package called Identification and Command (IDCOM), has been applied to the multivariable control of several refinery and petrochemical processes for the last ten years. But the detailed contents of MAC is not known, and there are several problems to be solved for more practically useful applications. The study discusses a general mathematical programming framework for multivariable MAC and a design method of MAC controllers. Simulation results show that MAC yields an improved performance over conventional control systems in terms of commonly used classical measures of response characteristics, including speed of response, settling time, cumulative error, and robustness. Control and state variable constraints are often overlooked in the development of control laws. Although recent developments with Dynamic Matrix Control (DMC0 and Quadratic Dynamic Matrix Control (QDMC) have addressed this problem, handling of constraints in the framework of MAC has not been addressed in the open literature. In this paper, a constrained multivariable version of MAC controller is presented, where Quadratic Programming (QP) is used to determine the optimal adjustments in the manipulated variables while satisfying constraints representing valve saturation and operational goals. Simulation results show that the proposed method yields a good performance over the traditional MAC strategy. The problem for non-minimum phase (NMP) systems is discussed within the framework of the MAC. So far, existing methods are computationally ineffective, and include a theoretical error in the derivation of Riccati equation. In order to solve above problems a new ...