Optimal control of building energy systems with multiple energy sources using predictive model based control and reinforcement learning

In this paper the control of building energy systems with multiple energy sources and storages are analysed. The goal is to efficiently coordinate the energy production and energy distribution from different sources in order to minimize the overall energy consumption. Model predictive control (MPC) and reinforcement learning (RL) based control approaches are proposed and exemplarily applied to an energy system of a residential building with different renewable energy sources. Because of the binary control inputs of the energy system a nonconvex integer optimisation problem arises. In order to solve the problem efficiently we apply a combined optimisation method that is integrated into the model predictive controller. Furthermore, a reinforcement learning based approach is developed and compared to the MPC controller in detail. Both methods are able to decrease energy consumption and keep thermal comfort at the same time.