Demonstrating model-based reinforcement learning for energy efficiency and demand response using hot water vessels in net-zero energy buildings

In this paper, we present a reinforcement learning framework to improve energy efficiency of domestic hot water provision using air source heat pumps. Simulations carried out using data from 40 houses shows 10–15% energy reduction, depending on occupant behavior. In absolute terms, this accounts to an energy reduction of about 150 kWh/a per house. The framework is extended to real world control, with energy savings of 27% demonstrated in a house over more than three months. We also explore the potential of using the same framework to provide demand response to the electric grid and find it to be asymmetric, i.e. positive flexibility (or upward regulation) is much higher than negative flexibility.