Randomized trial demonstrates energy cost reduction in smart home systems, highlighting the effectiveness of a novel framework.
Key Points
This research aims to develop an intelligent framework for optimizing energy management in smart homes using advanced forecasting and reinforcement learning techniques.
Proposed D2PG-SHEMS framework integrates ensemble PV forecasting with deep reinforcement learning.
Uses Random Forest and Long Short-Term Memory networks for accurate PV power predictions.
Utilizes Deep Q-Network for optimizing ensemble parameters and Deep Deterministic Policy Gradient for energy management control.
Reduces energy costs by 69.7% compared to RBC-based methods and 34.6% compared to standard reinforcement learning methods.
Maintains a comfort violation level of 35.190 during testing over a 72-hour period.