Optimization strategy based on deep reinforcement learning for home energy management

• Main Authors: Liu, Yuankun, Zhang, Dongxia, Gooi, Hoay Beng
• Other Authors: School of Electrical and Electronic Engineering
• Format: Article
• Language: English
• Published: 2021
• Subjects: Engineering::Electrical and electronic engineering; Deep Reinforcement Learning; Demand Response
• Online Access: https://hdl.handle.net/10356/148706
• Institution: Nanyang Technological University

With the development of a smart grid and smart home, massive amounts of data can be made available, providing the basis for algorithm training in artificial intelligence applications. These continuous improving conditions are expected to enable the home energy management system (HEMS) to cope with the increasing complexities and uncertainties in the enduser side of the power grid system. In this paper, a home energy management optimization strategy is proposed based on deep Q-learning (DQN) and double deep Q-learning (DDQN) to perform scheduling of home energy appliances. The applied algorithms are model-free and can help the customers reduce electricity consumption by taking a series of actions in response to a dynamic environment. In the test, the DDQN is more appropriate for minimizing the cost in a HEMS compared to DQN. In the process of method implementation, the generalization and reward setting of the algorithms are discussed and analyzed in detail. The results of this method are compared with those of Particle Swarm Optimization (PSO) to validate the performance of the proposed algorithm. The effectiveness of applied data-driven methods is validated by using a real-world database combined with the household energy storage model.