Air conditioning energy consumption accounts for over 58% of total household energy usage. As the most widespread cooling devices, residential air conditioners (RACs) therefore play a critical role in urban energy systems. This study collected the operational data from 496 RACs via an IoT platform. Through correlation analysis, the SHapley Additive exPlanations (SHAP) and physical principles, the average duration time and outdoor air temperature were identified as the key indicators influencing the regional energy consumption of RACs. Subsequently, the devices were categorized into three distinct clusters by applying k-means clustering to their daily duration time and the night proportion. These clusters represent daytime (15%), night (61%) and whole day running pattern (24%). To integrate the effects of the thermal environment and occupants’ behaviour, two regional energy factors (REF IoT and REF Climate ) were formulated using the key indicators of the duration time and temperature difference. These indicators were then used to develop regression models for predicting the regional energy consumption of RACs. The prediction models demonstrated strong performance, with those based on REF IoT achieving R 2 values above 0.89. Through applying the framework and indicators to dataset of 2024, the performance was similar to the models of 2023.
Dai et al. (2026) studied this question.