Neural-assisted HVACs optimal scheduling for renewable energy communities

Renewable energy communities (RECs) show a high potential for the efficient use of distributed energy technologies at local levels according to the Clean Energy Package of the European Union. Nevertheless, many problems are encountered during the planning phase of a REC because a large number of decision variables should be considered depending on the types of community participants and their load profiles. Moreover, the load schedule is fundamental with the aim of maximizing the energy shared within the community and the economic revenue of the REC. This is the main objective of this work that deals with the problem of optimal scheduling for electrical heating and cooling systems in a university lecture building included in a renewable energy community. The conditioning systems are significant electrical loads that can be scheduled both to reduce the energy requirements of the buildings and to synchronize energy injection and withdrawal from the grid, achieving energy self-consumption. The energy demand for the building is estimated using a specific simulation environment (EnergyPlus) which is experimentally validated with a campaign of measurements. From this software, a training dataset is created and used to implement a computationally lean neural network to represent through a black-box model the thermal-electrical behavior of the lecture building. Through the use of this neural model, the building thermal set points are processed by an optimization algorithm to maximize the economic revenue of the REC, while retaining the comfort boundaries of the building. The innovative proposed strategy is applied on a yearly simulation of the REC resulting in reduced energy required for thermal conditioning, higher cash-flows associated with the energy market and a smaller carbon footprint of the community.