Real-time energy scheduling applying the twin delayed deep deterministic policy gradient and data clustering

Official URL: https://dx.doi.org/10.1109/JSYST.2023.3326978

Smart homes are structural parts of the smart grid, since they contain controllable devices and energy management systems. In this work, we propose a reinforcement learning (RL)-based method for the energy scheduling of a smart home's energy storage system, heating ventilation and air conditioning system, and electric vehicle (EV). The proposed method targets to jointly minimize three evaluation metrics; the smart home's electricity cost, the residents' thermal discomfort, and the EV user's range anxiety. An advanced reinforcement learning algorithm, the twin delayed deep deterministic policy gradient (TD3), is utilized for this purpose together with a process, which is based on data clustering, that augments the similarity degree between the train and the test sets. As a result, the considered evaluation metrics show a significant improvement. The smart homes electricity cost, for instance, can be reduced by up to 11.2%, when compared with other RL-based works in the existing literature.