Conference Agenda

The automatic control system of air conditioners has recently gained attention as an effective approach to reduce energy consumption in buildings. The primary goal of these systems is to achieve both energy efficiency and thermal comfort for occupants. However, if the system fails to ensure occupant satisfaction even with energy-efficient operation, it risks being underutilized and losing its intended effectiveness. Therefore, for successful implementation, these systems must minimize user intervention by providing thermal comfort tailored to individual preferences. Conventional systems predominantly rely on PMV-PPD-based thermal comfort models, but these models often face limitations in real-world applications due to restricted parameters available for data collection and the discrepancy between individual thermal preferences and PMV-based statistical predictions. To address these challenges, this study proposes a method to derive personalized comfort ranges using data that can be practically collected on-site, including air conditioner control histories. Through experiments, user control patterns in specific thermal environments (temperature and humidity) were collected and utilized to develop both rule-based and machine learning-based models. To validate the models, occupant thermal preferences were surveyed and compared with the comfort ranges predicted by the models to assess their reliability. This study presents a data-driven approach for developing occupant-specific thermal comfort models, contributing to the practical implementation and advancement of autonomous air conditioner control systems.

As net-zero emerges as the critical issue, electrification is becoming important issue in building sector as well. Furthermore, the cooling energy demand is sharply increasing due to the impact of global warming. To utilize expensive renewable energy and to meet the increasing demand, the need for the energy efficiency and optimization in cooling system is significant technology. Therefore, this paper presents a reinforcement learning model for the residential cooling system, which can adapt to the residential environment within several days of observation (of occupant control of thermostats). The model utilizes obtained occupant control data and the energy consumption data from the electricity meter attached to the air conditioner with the neural network model to establish reasonable simulation-based training environment for the reinforcement learning agent, enables the model can be applied in existing building quickly while it maintains the ability to dynamically adapt to the occupant preference. The proposed model was validated in the chamber experiment, showed the energy saving witout sacrificing the thermal preference of the occupant, proved with the sensation vote and the preference vote result.

Integration of personalized human preferences in building control is vital for improving human comfort and satisfaction with indoor conditions. However, implementation of preference-based control in real buildings is very limited. This work presents the field implementation of preference-based thermal control in real offices, highlighting the impact of model predictive control (MPC) and low-cost local sensing. After developing thermal preference profiles based on experiments in private offices, preference-based HVAC control was integrated into the building management and control system. The results show (i) the potential thermal comfort penalty with conventional thermostat operation and (ii) the benefits of personalized MPC (dynamic setpoints) on energy use and personal comfort using local sensing. Depending on the thermal preference characteristics and climate conditions, preference-based predictive control showed 28-35% energy savings compared to conventional approaches and improvement in thermal comfort management.