Identifying robust transition pathways under unexpected events using Modeling to Generate Alternatives and Multi-Armed Bandits

Details

Supervisors

Contino, Francesco ; Kchaou, Mahdi

Faculty

Ecole polytechnique de Louvain

Degree label

Master [120] : ingénieur civil en mathématiques appliquées, à finalité spécialisée

Abstract

enThe urgency of climate change has placed the energy transition at the forefront of policy agendas worldwide, driving the development of optimisation-based energy system models as decision-support tools. Yet cost-optimality alone offers a fragile basis for long-term planning. Low probability but high-impact events, ranging from import disruptions and geopolitical shocks to technological setbacks and shifts in public acceptance, are systematically underrepresented in conventional energy planning, despite their demonstrated capacity to derail transition pathways. A planning framework that accounts for such contingencies is therefore necessary to yield strategies that are not merely optimal, but genuinely resilient. This thesis addresses this gap by proposing an integrated multi-horizon frame work combining two methodological pillars. The Modelling to Generate Alternatives approach is applied sequentially across transition phases to map the near-optimal solution space and identify structurally distinct trajectories, which are subsequently organised into a decision tree via k-means clustering. The robustness of these candidate pathways is then assessed through a Multi-Armed Bandit algorithm with an Upper Confidence Bound policy, enabling an efficient balance between the exploration of untested strategies and the exploitation of high-performing ones under a wide range of simulated unexpected events. The objective is to identify transition pathways that remain both economically acceptable and robust when disruptive events occur. Applied to the Belgian energy system, the framework identifies a portfolio of robust transition pathways rather than a single optimal trajectory. The results suggest that robustness is favoured by diversified low-carbon supply structures, where different production and flexibility levers can compensate for one another under stress. In particular, pathways combining substantial solar and nuclear deployment tend to perform well because they rely on structurally different sources of low-carbon capacity. Conversely, pathways depending too strongly on a single externally constrained supply option are more exposed to disruption. Overall, the analysis shows that several pathways within the near-optimal cost envelope can display markedly different robustness under unexpected events. Energy planning should therefore compare portfolios of near-optimal alternatives not only on cost, but also on their ability to remain feasible and effective under disruptive futures. This framework could be extended to additional decision horizons, broader sets of unexpected events, and more detailed operational representations of the energy system.