Deep Reinforcement Learning for Occupancy-Based Energy Management in Residential Demand Response

This paper presents a framework based on reinforcement learning that dynamically manages residential loads and heat pump consumption using price signals. The goal is to reduce the demand peaks generated in the system. A data-driven occupancy detection algorithm is incorporated to accurately capture occupant behavior and leverage this information to manage demand during times of absence, without compromising user comfort. This model allows the controller to intelligently adjust energy usage based on whether the building is occupiedAs a result, this solution enables flexible load shifting and demand response while preserving user comfort. Simulations with measured data demonstrate the ability of the proposed method to reduce energy bills while maintaining a desired comfort range. The significant difference in results when occupancy restrictions are actively applied compared to when they are ignored underscores the importance of occupancy-aware control strategies.