Advances in Industrial Model-Predictive Control.

Abstract

Due to the economically sensitive condition of the chemical and petroleum industries, we can no longer afford to operate with the inefficiencies of the past. Over the last 15 years we have found that on-line optimization implemented with model-predictive control recognizing process and operating policy conatraints, provides the best means for achieving the profit potential of our plants. Only model-predictive controllers permit the flexibility required to handle the constantly changing performance criteria, in particular the enforcement of operating constraints. However, the performance criteria of today's problems are becoming harder to quantify, while optimization systems are driving the processes over a wider range of operating conditions than ever before. Therefore, there is a need to improve the model-predictive control techniques so that practical performance criteria based on engineering judgement can be transparently specified, and that model inaccuracies are considered explicitly in the problem formulation. In this paper the state-of-the art in industrial model-predictive control is presented. An attempt is made to propose a path of evolutionary development in process control that will converge to a Unifed Theory replacing many of the ad- hoc solutions developed over the last thirty years. New techniques for multi-objective optimization and robust control are described that have the potential to allow us to improve the current technology in order to solve the control problem at hand. It is concluded that the complex control problems of today can only be solved through the development of a Unifed Theory along the concepts of model-predictive control. This Unified Theory of process control will then allow for the application of the Integrated Technologies of process optimization and control.