Data-Driven Estimation of Occupant Thermal Preferences from Air Conditioner Control Behavior in Residential Units

Autonomous control of building cooling systems represents an effective strategy to reduce energy consumption by preventing excessive cooling, thereby achieving both energy efficiency and thermal comfort for occupants. Conventional autonomous control methods have typically relied on fixed setpoints determined by PMV-PPD-based thermal comfort models. However, thermal preferences differ substantially among individuals, particularly in residential environments where personal characteristics strongly influence system operation, which complicates the application of statistical comfort models. Accounting for these differences would require extensive monitoring of both personal and environmental parameters, but such an approach is difficult to implement in practice due to the complexity of monitoring infrastructure and concerns over privacy. To address these challenges, this study proposes a method that derives occupant-preferred thermal conditions from cooling device control histories data, thereby minimizing the need for additional sensing parameters. Environmental data and operation logs of air conditioners were collected from each residential unit, and an artificial neural network (ANN) model was developed to predict household-specific preferred temperatures. For each residential unit, the upper boundary of the derived preferred temperature range was adopted as the setpoint for autonomous air conditioner control. Case studies conducted on four residential units demonstrated that, compared with manual control, the proposed method reduced energy consumption while maintaining comparable levels of occupant satisfaction. This study presents a data-driven framework for developing residential unit-specific thermal preference models and contributes to the practical implementation and advancement of autonomous air conditioner control systems.