Real-time energy management of a smart home based on deep deterministic policy gradient

In this paper, we develop a reinforcement learning-based scheme for the real-time energy management of a smart home that contains a photovoltaic-energy storage system. The objective of the proposed scheme is to minimize the electricity supplying cost by appropriately scheduling the storage system on a daily basis. The problem is formulated as a Markov decision process, which is optimized using the Deep Deterministic Policy Gradient (DDPG) algorithm. The main advantage of our proposed method compared to optimization-based ones is the ability to obtain effective daily schedules without relying on stochastic variables’ forecasts. In addition, nonlinearities related to the system’s operation can be effectively modeled, without the necessity of approximating them, as in linear optimization-based methods. The results confirm the ability of the DDPG agent to learn from historical data, and then to generalize the obtained knowledge to deal with real-time situations.