Structure-exploiting numerical methods for tree-sparse optimal control problems

PhD defense

Dimitris Kouzoupis

University of Freiburg

Monday, January 28, 2019, 14:00

Room 101-02-016, Georges-Koehler-Allee 101, Freiburg 79110, Germany

The main goal of this thesis was to explore new algorithms that enable the use of optimization-based techniques in control applications of fast dynamical systems. This task becomes especially challenging when uncertainty in the problem needs to be taken explicitly into account, increasing the size and complexity of the underlying optimization problems. We focus on cases where this uncertainty can be described by scenario trees and devise efficient numerical methods to solve such tree-sparse optimization problems efficiently. In this context, the numerical methods themselves are of course essential but the availability of efficient implementations is of equal importance. Therefore, considerable efforts have been spent to develop fast, embeddable, open-source software with some existing but also many new algorithmic features.