5048 Copyright © 1996 IFAC 13th Triennial World Congress. San Francisco. USA 3b-04 1 TUNING AND AUTO-TUNING IN PREDICTIVE CONTROL K. Yamuna Rani and H. Unbehauen Control Engineering Laboratory, Faculty of Electrical Engineering,Ruhr-University D-447BO Bochum,Germany email: office@esr.ruhr-uni-bochum.de Abstract: Several tuning guidelines proposed in literature have been investigated with the help of simulation studies for model predicitve control. Two typical examples are chosen as a third-order system with variable time constant and a second-order plus dead-time system with variable dead-time and time constant. The results illustrate the usefulness of the exist- ing methods. A modified version of a supervisory performance tuning procedure has also been applied to these systems with the intention of exploring the possibility of application of auto-tuning in Model Predictive Control. A new tuning procedure is finally presented on the basis of the results obtained using several previously existing tuning guidelines. Keywords: Predictive control, tuning characteristics, autotuners. 1. INTRODUCTION During the last decade, there has been an increasing trend in the industry towards the use of long range predictive controllers such as Dynamic Matrix Control (Cutler and Ramaker, 1980), Model Algorithmic Control (Richalet et aI., 1978), Generalized Predictive Control (Clarke et aI., 1987) and so on. Kramer and Unbehauen (1992) provide an extensive review concerning the theoretical and practi- cal aspects of several long range predictive controllers. In this study, attempts are made to compare several existing tuning guidelines for predictive controllers in order to arri- ve at a best tuning procedure with the help of a few simu- lation studies. Also, the concept of auto-tuning of the con- troller parameters after every setpoint change is examined for the same examples. In the next section, the prelimina- ries of predictive controllers are described briefly and the tuning guidelines proposed for Dynamic Matrix Control (DMC) and Generalized Predictive Control (GPC) are pre- sented. The following section presents the results of these methods when applied to typical simulation examples with process parameter changes such as lime constant, time delay, etc. Also, the results of application of a modified version of an auto-tuning method for GPC of these systems is discussed. Finally, the conclusions are presented. 2. TUNING METHODS IN PREDICTIVE CONTROL The concept of predictive control has been introduced mainly in order to deal with plants with complex characte- restics such as inverse response, dead time, etc. These controllers employ a common procedure: predicting the process outputs over a horizon based on the past input and output values using a model; computing a sequence of future manipulated input variable moves such that the predicted output response exhibits certain desirable cha- racterestics; implementing only the first control action of the sequence computed; and repeating the same procedure at the next sampling instant. Predictive control approaches have been classified into two major categories depending on the model employed - non