Stationary battery energy storage systems and electric vehicles become more and more popular at households with local photovoltaic generation. Besides improving self-consumption and autarchy, these batteries can provide flexibility to an external utility. Thereby, generation and demand uncertainty, as well as cost optimality, need to be considered when utilizing distributed flexibility.This paper discusses long short-term memory neural networks for photovoltaic generation forecast and persistence models for household load forecast with respect to their applicability in local energy management system optimization. Furthermore, a mixedinteger linear program is proposed to optimally utilize local flexible loads and storage systems. Its solution space yields the flexibility potential, which can be aggregated at flexibility pools. In order to disaggregate flexibility requests to a pool of distributed energy management systems, we propose a heuristic algorithm that can among others minimize the overall flexibility cost or maximize probability of flexibility delivery. The forecast models, the mixed integer linear program and the flexibility disaggregation are evaluated on realistic household photovoltaic and load profiles to demonstrate the full chain from local forecast to flexibility disaggregation under forecast conditions. Our experiments with flexibility disaggregation show that the probability to provide flexibility should not be neglected when it comes to distributed energy management optimization based on forecast models. CCS CONCEPTS• Hardware → Smart grid; • Computing methodologies → Uncertainty quantification; Neural networks.

No abstract