Individual room air-conditioning control in high-insulation residential building during winter: A deep reinforcement learning-based control model for reducing energy consumption

• Published in: Energy and buildings Vol. 323; p. 114799
• Main Authors: Sun, Luning, Hu, Zehuan, Mae, Masayuki, Imaizumi, Taiji
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 15-11-2024
• Subjects: Deep reinforcement learning; DQN; Overheating; Residential building; Room air-conditioner; DQN; Deep reinforcement learning; Residential building; Room air-conditioner; Overheating
• ISSN: 0378-7788
• DOI: 10.1016/j.enbuild.2024.114799

In recent years, the thermal insulation performance of residential buildings has been enhanced to reduce energy consumption. However, this enhancement often leads to air conditioning systems operating under ultra-low load conditions for extended periods, especially in individual rooms, which frequently results in sustained low efficiency. Additionally, during the winter, rooms tend to overheat due to the influence of solar radiation. In this study, we developed a deep reinforcement learning-based real-time & prediction full-stack control model, which can automate user-end air-conditioning control through home energy management system (HEMS). In this model, weather forecasts are utilised to mitigate overheating caused by solar radiation, thereby reducing energy consumption. Additionally, it can enhance the COP of air conditioners in low-load domains. Our successful empirical results indicate that the implementation of this model can reduce energy consumption by approximately 40% in winter.