Autoencoders for Hourly Load Profile Reconstruction in Renewable Energy Communities

The a-priori economic and energetic design of a Renewable Energy Community (REC) requires hourly electric generation and load profiles for the community members. Energy generation profiles are easily inferable, especially in the case of photovoltaics. Energy consumption trends, on the other hand, are more unpredictable. For this reason, the reconstruction of simulated hourly load profiles becomes a relevant area of study. In this paper, we propose a novel strategy to generate sensible hourly consumption profiles from information commonly found in energy bills, adopting a machine learning approach based on autoencoders. The results show that the proposed solution allows the generation of realistic hypothetical hourly load profiles.