 |
Universität Augsburg
|
|
|
Universität Augsburg
|
|
M. Sc. Johannes R. Kager
TUM Campus Straubing
spricht am
Dienstag, 13. Januar 2026
um
15:45 Uhr
im
Raum 1009 (L1)
über das Thema:
| Abstract: |
|
We consider two-stage robust mixed integer programming problems with finite uncertainty sets, where a subset of the decision variables - referred to as the first-stage variables - is shared across all scenarios. Assuming a minimization problem, the goal is to determine the values of these first-stage variables such that the worst-case objective value among all scenarios is minimized. In the single-objective case, the resulting min-max-min problem can be reformulated to a single-stage minimization MIP by introducing an auxiliary real variable and then solved with column-and-constraint generation methods. In the first part of the talk, we present an improved such method and apply it to a robust location routing problem and a robust berth allocation problem. In the second part, we consider the multi-objective setting, where optimizing the worst-case nondominated set introduces additional complexity. Depending on the choice of the quality indicator for evaluating the nondominated set, an equivalent reformulation into a single-stage MIP is often not possible anymore. In our work, we approximate the hypervolume indicator - the most-studied quality indicator in the multi-objective optimization literature - using a stripe-based subdivision of the objective space, which allows a reformulation as a single-stage MIP. We detail this formulation and present acceleration techniques to improve computational efficiency. We test our method on two-stage robust multiobjective versions of the selection problem and the assignment problem. |
| Hierzu ergeht herzliche Einladung. |
| Prof. Dr. Elisabeth Gaar |