IMPLEMENTATION OF CONSTRAINED PREDICTIVE OUTER-LOOP CONTROLLERS: APPLICATION TO A BOILER CONTROL SYSTEM Fernando Tadeo, Teresa Alvarez Departamento de Ingenieria de Sistemas y Automática Facultad de Ciencias. Universidad de Valladolid. 47011 Valladolid, Spain fernando@autom.uva.es, Tlf:+34 983 423566, Fax:+34 983 423161 Mike Grimble Industrial Control Centre, Strathclyde University, United Kingdom Ramon Vilanova Automatic Control and Systems Engineering Group, Dept. of Computer Science, Universitat Autonoma de Barcelona, 08193 Bellaterra – Barcelona, Spain Abstract: This paper addresses the problem of implementing predictive controllers for supervisory level control systems. In this configuration the manipulated variables calculated by the Predictive Controller are used as command signals for the Distributed Control Systems, which provide references to the operator-tuned local PID controllers that act on the physical system. This structure introduces the problem of loosing of performance if the inner-loop controllers are re-tuned. The paper discusses the solution to this problem based on the use of a two-degrees-of-freedom structure in the inner loop, that separates open and closed-loop properties. Both design guidelines and robustness issues are discussed. Copyright © 2002 IFAC Keywords: Supervisory Control, Predictive Control 1. INTRODUCTION For multivariable process control problems with strong interactions between the controlled and manipulated variables and strict constraints, the conventional multiloop PID control configuration may not provide adequate control performance (Seborg, 1994). Model-Based Predictive Control (MBPC) solves these problems by predicting future process behaviour and calculating control variables taking into account the process constraints (Clarke et al. 1987). These techniques have been applied very successfully to different Process Control Problems (Camacho and Bordons 1995, Froisy 1994). A common structure for a MBPC in industrial process control problems is shown in figure 1. The MBPC calculates the future control signals based on the measured variables. These control signals (manipulated variables), are sent to the Distributed Control System as command signals for the actuators (such as valve positioning commands). Local PID controllers act on the physical system to obtain the desired manipulated variable, which is fed-back to both the local PID and the predictive controller. This is one of the structures discussed by Lee et al. (1997), where it is called a “Cascade control – series connected system”. It was also studied by Saez et al. (2000), where it was proved that under certain conditions the master controller could be selected to make control characteristics independent of the slave controller. However the solution proposed by Saez et al. (2000) is not used in this paper, as it is based on perfect knowledge of the slave-loop controller (which is not always possible), and generates a pole/zero cancellation between slave and master controllers (which in certain cases is undesirable). It must be pointed out that direct control of the plant by the predictive controller is also frequent in practical implementations (an example for a Steam Generator is presented in Khotare et al., 2000). However, plant operators in industry are not usually ready to permit direct control of the plant by they predictive controllers, unless they are already very familiar with predictive control. The implementation discussed in this paper makes possible to prove the improvement in performance of predictive controllers, which could later be upgraded to direct control, if desired. In most industrial implementations these PID loops in the slave loop are tuned by the operators of the plants, based on their knowledge of the local process. The Predictive Controller includes, in the models used for prediction, the PID, actuator and measurement filter dynamics. That means that if any of the inner-loop controllers, which will be referred to as slave controllers, is re-tuned the real system differs from the model used for prediction. This difference causes Copyright © 2002 IFAC 15th Triennial World Congress, Barcelona, Spain