Modelling of a multi-stage energy management control routine for energy demand forecasting, flexibility, and optimization of smart communities using a Recurrent Neural Network

This paper proposes an innovative algorithm for community energy management control, able to involve customers in energy trading by exploiting their potential energy flexibility. The main innovation relies on a matrix-based control system where the strategy considers individual and community priorities simultaneously. Through the individual energy flexibility and the community energy pool, the aggregated network energy supply is controlled and shaped. The presented model presents a generalized structure based on control volumes, and it can be universally applied to energy communities of different sizes, number of participants, energy carriers, penetration of photovoltaics, and electric vehicles. The predictive system is conceived from a recurrent neural network, which performs a real-time prediction on energy demands in buildings. Suitable energy flows optimization is also presented with different implications for economic and energy savings. Finally, to show the potential of the developed model, a suitable case study analysis is presented. Important results include the achievement of a typical win-win condition, where both the distribution system operator and final customers benefit from this strategy. Specifically, a reduction of energy demand during demand response events of about 21% is achieved, whereas the interaction with the electricity network decreases of about 15%.