Energy system optimization models (ESOM) are powerful tools for improving decision making in the transition towards carbon-free energy systems.

The results provided by ESOMs are greatly influenced by data uncertainty since they are considered on a future time horizon. For instance, the possible evolution of energy prices, energy production and demand or the efficiency of technologies must be taken into account. Although many works have started in recent years to study the impact of these uncertainties on the solutions, it has been pointed out that modeling simplifications may induce significant bias in the obtained results.

The work proposed in this new PhD topic aims at studying the response of ESOM along energy system design and transformation steps, and reducing or assessing the impact of uncertainties as early as possible in the process. It will especially aims at limiting the bias related to model simplification, by systematically propagating relevant information from more detailed models towards simplified models used for sensitivity analysis and optimization under uncertainty. To this aim, the currently envisioned path is to leverage techniques such as machine learning, and in particular the constraint learning approach, to extract relevant information from simulation and inject back into the simplified optimization models.

As a result, the work is expected to improved the methods currently in use for designing and improving energy systems at local level, in order to favor energy savings, and limit CO2 emissions as well as other environmental impacts.