Real-time optimization under uncertainty: Accelerated and distributed modifier-adaptation schemes

René Schneider, EPFL Lausanne


The desire to operate technical processes in a safe and economically optimal way has motivated the development of various real-time optimization methods. A key challenge for the computation of optimal inputs or setpoints is structural plant-model mismatch. So-called modifier-adaptation schemes have been shown to be effective in compensating for such mismatch by incorporating process measurements. In this talk, we present recent theoretical advances of modifier-adaptation algorithms. The focus will be on modifications that can accelerate convergence or enable distributed implementations for interconnected systems. Important application aspects such as privacy of local models will also be discussed. The theoretical developments will be illustrated by simulated examples from chemical engineering.