Transportation activities are among the main sources of carbon emissions. In this context, electric vehicles (EVs) have emerged as a greener alternative to internal combustion vehicles due to their zero-emission nature during operation. However, the rapid growth of EV adoption poses a major challenge to the power grid's ability to meet increasing demand from renewable sources. This thesis addresses this challenge by recognizing two interdependent decision levels---strategic and operational---forming a chicken-and-egg dilemma. The strategic level concerns the deployment of charging infrastructure (location and capacity), while the operational level focuses on the assignment of EVs to charging stations for a given energy price. Each level depends on the configuration of the other, involving multiple stakeholders, notably EV users and charging facility operators (CFOs). We model this interaction as a hierarchical decision-making process. From a mathematical programming perspective, this action--reaction dynamic is formulated as a bilevel optimization problem, where the leader (CFO) anticipates the reaction of the follower (EV users) within its decision process. Despite growing interest in bilevel optimization for this scope, the research direction remains fragmented. This thesis provides three contributions to address the chicken-and-egg dilemma within a coherent methodological framework. First, we present a comprehensive review of bilevel models applied to charging infrastructure management. A proposed taxonomy structures the research field and identifies promising extensions. The analysis reveals that, in urban contexts, coordination of energy supply is a more critical issue than the availability of charging points. Second, we introduce the textit{Charging Facility Location-and-Operation Problem} (CFLOP), a bilevel optimization model integrating a novel network augmentation approach called textit{wind-flow-chain} (WFC). This approach captures both EV driving behavior and their assignment to charging stations. The lower level represents users' path selection minimizing generalized travel cost (driving, waiting, charging time, and energy cost), while the upper level determines the location and capacity of stations under budget and design constraints, maximizing CFO revenues. We define equilibrium conditions involving a continuous linking variable for energy price and propose three single-level reformulations under bounded and unbounded rationality. Computational experiments and sensitivity analyses demonstrate the model's efficiency in supporting strategic and operational planning. Third, to reduce the computational burden, we develop a Benders decomposition approach to solve the CFLOP. Based on the strong-duality single-level reformulation, the selected partition yields a feasibility subproblem incorporating the dual multipliers of the lower level, separable by EV user and enabling a multi-cut scheme. Numerical results show that the Benders decomposition outperforms direct single-level solutions from standard mixed-integer programming solvers, achieving small optimality gaps in much shorter times. Overall, this thesis proposes a unified methodology combining bilevel modeling and Benders decomposition to analyze the planning and operation of EV charging infrastructures under the energy transition, offering a robust decision-support framework for policymakers and infrastructure operators.
Mme Luce BROTCORNE Directrice de recherche Université de Lille Directrice de thèse, Mme Ivana LJUBIC Full professor ESSEC Business School Rapporteure, M. Michel GENDREAU Full professor Polytechnique Montréal Co-directeur de thèse, M. David REY Full professor SKEMA Business School Rapporteur, Mme Hanane DAGDOUGUI Full professor Polytechnique Montréal Examinatrice, M. Walter REI Full professor Université du Québec à Montréal Examinateur, M. Miguel F. ANJOS University of Edinburgh Invité.
Thesis of the team INOCS defended on 02/02/2026
français
English