Design Methodology of Modifier Adaptation for On-Line Optimization of Uncertain Processes

This paper is concerned with the on-line optimization of uncertain processes using the modifier-adaptation technology. In this approach, an optimization model is solved repeatedly and the available measurements are used to correct the values and gradients of the predicted outputs in that model. Following the paradigm of dual control, a successful modifier-adaptation scheme must reconcile two conflicting objectives, namely optimizing the process and obtaining accurate gradient information. During the on-line execution phase, a certain quality of the gradient estimates is enforced through additional constraints in the optimization problem. Then, a systematic off-line procedure is developed for the design of these constraints.