Anmerkungen:
Diese Datenquelle enthält auch Bestandsnachweise, die nicht zu einem Volltext führen.
Beschreibung:
This thesis deals with the distributed optimization of constraint-coupled systems. This problem class is often encountered in systems consisting of multiple individual subsystems, which are coupled through shared limited resources. The goal is to optimize each subsystem in a distributed manner while still ensuring that system-wide constraints are satisfied. By introducing dual variables for the system-wide constraints the system-wide problem can be decomposed into individual subproblems. These resulting subproblems can then be coordinated by iteratively adapting the dual variables. This thesis presents two new algorithms that exploit the properties of the dual optimization problem. Both algorithms compute a quadratic surrogate function of the dual function in each iteration, which is optimized to adapt the dual variables. The Quadratically Approximated Dual Ascent (QADA) algorithm computes the surrogate function by solving a regression problem, while the Quasi-Newton Dual Ascent (QNDA) algorithm updates the surrogate function iteratively via a quasi-Newton scheme. Both algorithms employ cutting planes to take the nonsmoothness of the dual function into account. The proposed algorithms are compared to algorithms from the literature on a large number of different benchmark problems, showing superior performance in most cases. In addition to general convex and mixed-integer optimization problems, dual decomposition-based distributed optimization is applied to distributed model predictive control and distributed K-means clustering problems. ; Diese Arbeit befasst sich mit der verteilten Optimierung von Restriktions-gekoppelten Systemen. Diese Problemklasse tritt häufig in Systemen auf, die aus mehreren einzelnen Teilsystemen bestehen, die durch geteilte begrenzte Ressourcen gekoppelt sind. Das Ziel ist es, jedes Teilsystem verteilt zu optimieren und gleichzeitig sicherzustellen, dass die systemübergreifenden Restriktionen eingehalten werden. Durch die Einführung dualer Variablen für die systemübergreifenden ...