The transition towards carbon-neutrality of a whole-energy system (i.e. including all streams of energy carriers and demands) is uncertain. Therefore, instead of establishing single-shot plans towards 2050 (and beyond), i.e. perfect foresight, policy makers rather go through multiple rolling-horizon short-term decisions, i.e. myopic approach. Meeting the environmental objectives while minimizing the cost of the system, accounting for this decision-making process, the uncertainties, and potential shocks/crisis, require therefore a framework to assess the relevance and the timing of the decisions throughout the transition. For this purpose, a reinforcement-learning agent has been trained, interacting with its environment, a cost-based whole-energy system model(EnergyScope). Repeating the whole transition with different sequences of actions allows the agent to come up with a robust policy towards sustainability, considering the variation of the parameters of its environment. This framework has been applied to the case of Belgium. Given the ambitious CO2-budget target, the main lever of action is to limit the emissions and the consumption of natural gas as the cost-optimum would anyway get rid of oil products and keep on using as much coal as allowed.

Speaker : Xavier RIXHON