Home Energy Management System Based on Multi-Agent Deep Reinforcement Learning Handling the User'S Thermal Preferences

With the rising presence of digitalization, an increasing number of residents can now manage their smart appliances through home energy management systems (HEMS). However, HEMS faces significant challenges, including the complexity of user comfort modeling, various uncertainties in parameters, and the inherent difficulty of high-dimensional decision-making in dynamic environment. To overcome these issues, this paper proposes a data-driven framework for HEMS based on multi-agent reinforcement learning (MARL). In this model, a thermal comfort preference model is introduced to handle the user's uncertain behavior. Also, a long short-term memory (LSTM)-based machine is developed for accurately predicting photovoltaic (PV) generation time-series. Finally, after combining PV predictions with thermal comfort assessments, a deep-double Q-network (DDQN) algorithm is applied to achieve optimal decision-making in HEMS. Numerical experiments show that the proposed method effectively tackles the challenge of highdimensional integrated decision making under uncertainty.