A.R. Van Cauwenberghe

Extended Prediction Self-Adaptive Control

Abstract Extended Prediction Self-Adaptive Control is a control strategy in which the calculation of the controllers actions is based on an adaptive long-range prediction of the resulting process output. The forecast is made based on a black box model of the process dynamics. Its parameters are identified in real time. The technique seems to be quite robust with respect to modelling errors. Moreover the control objective function has a strong intuitive appeal to the process operator. Therefore the method is especially well suited for application to real-life control problems.

A self-tuning multistep predictor application

A self-tuning multistep predictor is presented. It predicts the output of a stochastic process with unknown, possibly slowly time-varying parameters over a range of several sampling periods in the future. At each sampling instant it is tuned by using a recursive least-squares parameter estimator in real time. By doing this, the combination predictor-estimator converges fast to the optimal predictor for processes with known parameters (self-tuning property). The method seems to have powerful capabilities as an aid in controlling complex industrial processes which are until now only operated under manual control. The predictor can be used by the operator in selecting an appropriate control action (decision making). A typical application, the control of a blast furnace, is extensively dealt with in the paper. The paper opens new perspectives in the domain of self-tuning controllers, and it has practical importance as is indicated by the blast-furnace experiment.

Typical Application Possibilities for Self-Tuning Predictive Control

Abstract Control based on self-tuning prediction has some typical application possibilities. Two such realizations are described in this paper. The first one deals with dead-time control in a heat transport process. A predicted value of the process output is fed back, leading to a compensation of the time-delay. This allows for higher loop gains, resulting in a better control performance as is demonstrated by the comparison with a PI-regulator. The second implementation of self-tuning predictive control is in a decision-making strategy used in an electrical servo. This application shows that very satisfactory control performance can be obtained with rather primitive control tools (on-off) combined with a self-tuning predictor. The results of the strategy are compared to those of well-known control methods, among which a PD-controller.

Simple Self-Tuning Multistep Predictors

Abstract The term multistep prediction is used to indicate the prediction of a stochastical ly disturbed dynamical signal over a longer range of future consecutive sampling instants. This technique is a keystone in some recently developed predictive control strategies with promising application possibilities. The prediction is based on a mathematical model of the dynamical process. When this model is numerically unknown, self-tuning multistep prediction is a valuable alternative. However, its computational burden is rapidly growing with the length of the prediction range. A simple and faster suboptimal multistep predictor is compared to the optimal one. The most attractive features are that the increase in prediction error due to the non-optimality can be computed in real-time from the self-tuners parameter estimations and that the suboptimal algorithm can be adapted such that this increase does not exceed a prescribed value. Its use for the design of simple self-tuning predictive controllers looks promising.

A Self-tuning Predictor as Operator Guide

Abstract A self-tuning multistep predictor is presented. It predicts the output of a stochastic process with unknown, possibly tiIre-varying, parameters over a range of several sampling periods in the future. At each sampling instant it is tuned by using a recursive least-squares parameter estimator in real time. By doing this, the combination predictor-estimator converges fast to the optimal predictor for processes with known parameters (self-tuning property). The method seems to have powerful caIXiliilities as an aid in controlling complex industrial processes which are until now only operated under manual control. The predictor can be used by the operator in selecting an appropriate control action. A typical application, the control of a blast furnace, is extensively dealt with in the paper. The paper opens new perspectives in the donain of self-tuning controllers, and it has practical importance as is indicated by the blast-furnace experiment.

Self-Adaptive Long-Range Predictive Control

A control strategy is presented in which the calculation of the controllers actions is based on a adaptive long-range prediction of the resulting process output. The forecast is made based on a black box mathematical model of the process dynamics. It is identified in real time. The technique of long-range predictive control seems to be quite robust with respect to modeling errors and the control objective function has a strong intuitive appeal to the operator.

Brief paper: Simulations of adaptive controllers for a paper machine headbox

The paper headbox is a crucial part of a paper machine, largely influencing the paper quality. Since this can be improved by better control, the evaluation of advanced control methods is certainly justified from an economical point of view. Not only do disturbances have to be rejected, but smooth startups and fast setpoint changes must also be possible. To design a good controller, some parameters need to be known, the determination of which can be cumbersome. However, this can be considerably alleviated by adding an on-line parameter estimator to the controller. This allows the controller to adapt itself to slowly changing process characteristics. In this paper different adaptive control methods are used on the simulated process and compared to the classical PI-controls. Particularly, a single-output self-tuning controller with feedforward action and a multivariable adaptive deadbeat controller are promising. The last one provides simultaneous offset-free control of both process outputs.

Application of several parameter-adaptive controllers to a paper machine headbox

Abstract Several authors have pointed to the importance of optimal paper headbox control as a crucial part of the global paper making process. External disturbances influence the output, also many startups of the process may be required. Since the paper quality can be improved by better control, the evaluation of advanced control algorithms is well justified. The introduction of process computers and microprocessors in the control environment has opened new perspectives for the control of paper making too. To design a good controller, a number of numerical coefficients must be known accurately. The determination of these parameters can be cumbersome. However, this can be considerably alleviated by adding an on-line parameter estimator to the controller. Then the regulator can adapt itself to slowly changing process characteristics. These parameter-adaptive controllers were extensively tested on the Laboratory of Automatic Control in Ghent. Several combinations of controllers and parameter estimators are possible. It will be demonstrated in the paper that several of these controllers improve the control quality of the paper headbox. They are not very sensitive to the noise structure and cause no offset of the output variable. Since the implementation of such adaptive controllers requires very few a priori knowledge, it may be expected that in the future many industrial applications of these will be seen.

Neural-network-based multiple feedback long-range predictive control

Abstract After introducing neural-network-based nonlinear long-range predictive control for processes with time-delay, the paper presents several on-line corrections of long-range predictive outputs and proposes neural-network-based multiple feedback predictive control for nonlinear cascade industrial processes. A variable correction coefficient and its design approaches are presented. The simulation experiments have shown that the algorithm is able to effectively overcome disturbances and improves both the static and dynamic performances of the controlled system.

Brief paper: A new Liapunov technique for stability analysis of chemical reactors

A new type of Liapunov function is proposed, for the simultaneous generation of regions of attraction of all stable equilibria of a chemical reactor system. The method is computationally simpler than the existing procedures, due to the fact that only one Liapunov function is required for investigating all singular states at once. The obtained results may be considerably improved by combining the proposed Liapunov function with some simple topoogical arguments.