How do people set air conditioning temperature setpoint in urban domestic–Behavior model in Chinese three climate zones based on historical usage data

Abstract

Predicting user behavior in setting air conditioning (A/C) can help develop automatic A/C control techniques, but it is challenging. Existing studies have largely focused on A/C switching behavior models, with fewer studies exploring temperature-setting behavior models in various climate zones. This study analyzed differences in household A/C temperature setpoint behaviors in three climate zones, using usage data collected from embedded sensors in A/C during the cooling season. Further, five machine learning (ML) algorithms were applied to model the A/C temperature setpoint behavior for the three climate zones. The results showed that the behavior is related to time, indoor air temperature (Tin), outdoor air temperature (Tout), temperature setpoint before settings (Tset'), and cumulative time the A/C is turned on (tc). A/C usage behavior varied significantly among users in different climate zones. Users in hot summer and warm winter (HSWW) have a higher acceptable Tin and the opposite in Cold zones. In HSWW zones, users were able to find their preferred setting temperature more quickly. By applying the random forest algorithm and proper data pre-processing, the model can predict the “Warmer”, “Keep setpoint unchanged” and “Cooler” settings with 0.700–0.810 macro F1 score and 0.936–0.961 accuracy. This shows the excellent performance of ML algorithms in predicting user behavior with large data sizes. The Tin, Tset', and tc are important input features. This work aims to model A/C temperature setpoint behavior through ML algorithms to help develop intelligent A/C control methods for different climate zones in the future.