Analysis and modeling of air conditioner usage behavior in residential buildings using monitoring data during hot and humid season

Abstract

With split air conditioner (AC) becoming one of the most popular thermal comfort-related appliances in residential buildings in China, many issues, such as high energy consumption, have arisen. One crucial way to address these issues is to study the usage of split AC models, which can be greatly affected by occupant behavior. In this study, we conducted a large-scale field measurement among 34 residential buildings in eight cities distributed in three climate zones using wireless sensors during summer. The AC usage patterns in the three climate zones were analyzed, and models based on artificial neural networks and gradient boosting decision tree algorithms were developed for predicting AC on/off behavior. The results showed that the average usage duration and energy consumption of AC per day in the three climate zones were 7–10 h and 2.78–3.00 kWh, respectively, and that there was a large difference among different households. The AC usage patterns of households in the three climate zones had certain similarities. The models developed in this study showed higher performance and accuracy than the logistic regression-based model, and indoor relative humidity and CO2 concentration were found to significantly improve predicting accuracy. The models can generate an AC operation schedule as an input boundary condition to improve the accuracy of residential building energy consumption simulations in future studies.