Real-Time Optimization Via Modifier Adaptation Integrated with Model Predictive Control

Abstract In order to deal with plant-model mismatch, real-time optimization schemes use some adaptation strategy based on measurements. The modifier-adaptation approach is to correct the constraints and gradients in the optimization problem by adapting the values of bias modifiers expressing the difference between the constraints and gradients of the plant and the model. The approach has the ability to converge to the plant optimum but does not guarantee feasibility prior to convergence if the evaluated optimal inputs are applied directly to the plant. In this paper, an approach for integrating modifier adaptation with model predictive control is presented where both automation layers use the same decision variables. The control targets are included as equality constraints in the real-time optimization problem, and are continuously enforced by the model predictive controller. In order to apply modifier adaptation, new gradient modifiers are defined that correct the projected gradients in the tangent space of the equality constraints, rather than the full gradients. An illustrative case study shows the applicability of the approach.