The 7th of December, 2016, Sezin Afsar succesfully defended her thesis titled Revenue optimization and demand response models using bilevel programming in smart grid systems.
The jury was composed by:
- Director: Luce Brotcorne, Researcher, Inria Lille-Nord Europe
- Co-director: Gilles Savard, Professor, Ecole Polytechnique Montréal
- Rapporteur: Roberto Wolfler-Calvo, Professor, Université Paris 13
- Rapporteur: Bernard Gendron, Professor, Université de Montreal
- Jury member: Dominique Quadri, Associate Professor, Université Paris Sud
- Jury member: Bernard Fortz, Professor, Université Libre de Bruxelles
- Jury member: Miguel Anjos, Professor, Ecole Polytechnique Montréal
- Invited member: Sandrine Charousset, Project Manager, EDF
Thesis abstract: This thesis is concerned with revenue optimization of an energy provider. A bilevel programming approach is proposed to model the relationship between the energy provider (leader) and power users (follower). The leader intends to achieve an optimal trade-off between revenue and peak load whereas the follower minimizes total cost of users to achieve system optimality.
A smart grid structure that allows two-way communication is assumed to interconnect users and to schedule their demand regarding the prices. Day-ahead real-time prices are read by each customer’s smart meter and the response is coordinated.
In this thesis, we propose several bilinear bilevel programs that are presented and reformulated as single-level mixed integer problems using the KKT conditions of the follower’s problem. These MIPs are solved to optimality for randomly generated instances using a commercial software. Different versions of the models are tested and compared.
In order to s olve large instances, several heuristics are developed. Two of these methods are shown to be efficient and solve large instances that cannot be solved within a reasonable time interval using exact method. Their outputs are compared to the exact solutions for small instances and their performances are evaluated.
Finally, we address the robust bilevel optimization problem, discuss existing approaches, give illustrative examples, and propose avenues for future research.
