| Keywords |
Deep Reinforcement Learning; Building Autonomous Control; Retail Facility; Thermal comfort; Energy Reduction |
| Abstract |
This study evaluated the field performance of an artificial intelligence-based autonomous building control (aiBAC) model in a large retail building during the summer season. The control model was developed using reinforcement learning and pre-trained in a hybrid simulation environment to ensure stable operation under real building conditions. A total of 50 valid operating days were analyzed, including 29 days under conventional control and 21 days under autonomous control. Energy performance was assessed using gas consumption, while thermal comfort performance was evaluated using the thermal compliance ratio and mean out-of-band intensity. The results showed that the autonomous control reduced daily gas consumption by 9.2% compared with conven- tional control. In addition, the temperature compliance ratio improved from 51.4% to 74.6%, while the mean out-of-band intensity decreased from 0.82°C to 0.55°C, indicating improved thermal comfort and more stable indoor temperature control. Further analysis under different outdoor temperature and occupancy conditions confirmed consistent energy-saving performance. These findings demonstrate that reinforcement learning-based autonomous control can simultaneously improve energy efficiency and thermal comfort in real-world retail buildings and has strong potential for practical application in intelligent building energy management |