Skip to Main content Skip to Navigation

Contributions to static and adjustable robust linear optimization

Abstract : Uncertainty has always been present in optimization problems, and it arises even more severely in multistage optimization problems. Multistage optimization problems underuncertainty have attracted interest from both the theoretical and the practical level.Robust optimization stands among the most established methodologies for dealing with such problems. In robust optimization, we look for a solution that optimizes the objective function for the worst possible scenario, in a given uncertainty set. Robust multi-stage optimization problems are hard to solve even heuristically. In this thesis, we address robust optimization problems through the lens of decompositions methods. These methods are based on the decomposition of the robust problem into a master problem (MP) and several adversarial separation problems (APs). The master problem contains the original robust constraints, however, written only for finite numbers of scenarios. Additional scenarios are generated on the y by solving the APs. In this context, heuristic solutions and relaxations have a particular importance. Similarly to combinatorial optimization problems, relaxations are important to analyze the optimality gap of heuristic solutions. Heuristic solutions represent a substantial gain from the computational viewpoint, especially when used to solve the separation problem. Because the adversarial problems must be solved several times, good heuristic solution may avoid the exact solution of the APs. The main contributions of this work are three-fold. First, we propose a new relaxation for multi-stage problems based on the approach named perfect information in the field of stochastic optimization. The main idea behind this method is to remove nonanticipativity constraints from the model to obtain a simpler problem for which we can provide ad-hoc combinatorial algorithms and compact mixed integer programming formulations. Second, we propose new dynamic programming algorithms to solve the APs for robust problems involving budgeted uncertainty, which are based on the maximum number of deviations allowed and on the size of the deviations. These algorithms can be applied to lot-sizing problems and vehicle routing problems among others. Finally, we study the robust equitable sensor location problem. We make the connection between the robust optimization and the stochastic programming with ambiguous probabilistic constraints. We propose linear models for several variants of the problem together withnumerical results.
Complete list of metadatas

Cited literature [87 references]  Display  Hide  Download
Contributor : Abes Star :  Contact
Submitted on : Tuesday, May 22, 2018 - 4:10:06 PM
Last modification on : Wednesday, July 4, 2018 - 4:44:03 PM
Long-term archiving on: : Monday, September 24, 2018 - 5:23:37 PM


Version validated by the jury (STAR)


  • HAL Id : tel-01797497, version 1



Marcio Costa Santos. Contributions to static and adjustable robust linear optimization. Other [cs.OH]. Université de Technologie de Compiègne, 2016. English. ⟨NNT : 2016COMP2312⟩. ⟨tel-01797497⟩



Record views


Files downloads