D2PG-SHEMS: Integrating Ensemble PV Forecasting and Deep Reinforcement Learning for Smart Home Energy Management Systems

Abstract Demand response has become an important part of smart home energy management, and an excellent demand response policy can be very effective in reducing energy consumption and comfort violation. However, the uncertainty of renewable energy, the complex interaction between photovoltaic (PV) systems, controllable equipment and the grid, and the uncertain behavior of users bring great challenges to the optimization of smart home energy management. In this paper, we propose an intelligent home energy management framework, Dual Predictive Control-Deep Deterministic Policy Gradient method for Smart Home Energy Management Systems(D2PG-SHEMS). Our method employs an ensemble model combining Random Forest (RF) and Long Short-Term Memory (LSTM) networks, where Deep Q-Network (DQN) is employed to optimize the ensemble parameters for accurate PV power forecasting. Based on the ensemble model's predictions, the control module utilizes Deep Deterministic Policy Gradient (DDPG) combined with Imitation Learning (IL) to enhance training efficiency for optimal energy management. According to the experimental results over a 72-hour testing period, the proposed method reduces energy costs by 69.7% and 34.6% compared to the RBC-based and standard RL methods, respectively, while maintaining the comfort violation level at 35.190.