Personalized Federated Deep Reinforcement Learning for Smart Home Energy Management

Deep reinforcement learning (DRL)-based energy management methods for smart homes have received wide attention due to their low reliance on thermal dynamics models and uncertainty parameter information. However, they mainly focus on utilizing local operational data to train energy management policies, which may lead to overfitting and reduced decision-making performance. Although federated deep reinforcement learning (FDRL) can be used to alleviate this problem, it is inefficient in the context of heterogeneous smart homes. To overcome this drawback, this paper proposes a personalized energy management approach for each heterogeneous smart home based on personalized federated deep reinforcement learning (PFDRL). The proposed approach consists of two stages. First, a global energy management strategy is derived using FDRL from the strategies of all heterogeneous smart homes. Then, a personalized fine-tuning mechanism based on distance constraints is designed to further adapt the global strategy into a personalized energy management strategy for each smart home. Simulation results show that the proposed approach achieves the best thermal comfort performance among all schemes and simultaneously reduces energy costs by 6.69%-15.70% compared with the rule-based scheme and the independent training scheme.