BaCaTeC. Data-Based Optimization in Real Time for Dynamic Systems
- Enrique Zuazua, Miroslav Krstic
- Supported by HighTech Research between Bavaria and California
- Duration: 2021 – now
Unlike most machine learning algorithms, which have yet to be equipped with guarantees of convergence and stability in real time for feedback applications to dynamical systems, one of the earliest example of data-based optimization algorithms, the so-called “extremum seeking” (ES) approach, whose original idea is traced back to a century-old patent in France in 1922, possesses provable properties of stability and even assignable convergence rates.
This is an example of Artificial Intelligence (AI) decades before the notion of AI was formalized. For nite-dimensional systems, the mathematical guarantees for ES were developed by Prof. M. Krstic around 2000. Over the last few years, he has extended the ES algorithm design and stability analysis from Ordinary Differential Equations (ODE) to Partial Differential Equations (PDE). The visit by Prof. M. Krstic to FAU will be an opportunity to explore the cooperation in this area with further PDE applications in view, including gas transport networks.