Smart home's energy management applying the deep deterministic policy gradient and clustering

Abstract

Smart buildings, equipped with controllable devices and energy management systems are expected to be substantial parts of the future energy grids. In this paper, a Reinforcement Learning (RL)-based method is developed for the energy scheduling of a smart home's energy storage system, which is also equipped with a photovoltaic system. The proposed scheme aims to minimize the electricity cost of the smart home; the overall problem is formulated as a Markov decision process, and it is solved by applying the Deep Deterministic Policy Gradient (DDPG). The main advantage of the proposed method is that increases the degree of similarity between the train set and the test set, through data clustering, achieving superior energy schedules than the existing RL-based approaches.