Artificial-neural-network-based model predictive control to exploit energy flexibility in multi-energy systems comprising district cooling

Abstract

District cooling systems (DCSs) belonging to multi-energy systems can be managed by model predictive controls (MPCs) designed to reduce the amount of electrical energy collected from the grid for backup cooling systems when there is a temporal mismatch between energy demand and availability. In this paper, a DCS recovering cold thermal energy from a liquid-to-compressed natural gas fuel station is used in an 8-user residential neighborhood to provide space cooling in summertime. In the residential neighborhood, there is a multi-energy system, including the DCS, photovoltaic panels, and backup systems based on variable-load air-to-water heat pumps. One user of the district was allowed to manage its energy demand with an MPC based on an artificial neural network (ANN). By integrating the ANN-based MPC routine in the building simulation environment and unlocking the energy flexibility of thermostatically controlled loads (TCLs) using variable setpoints, it was possible to reduce electrical energy consumption up to −71% with respect to a reference case with a rule-based control. This work highlights also the importance of the ANN training process for a proper representation of the TCL flexibility in the building model, which is not a trivial aspect to be taken into account in data driven models.