Characteristics analysis and modeling of occupants' window operation behavior in hot summer and cold winter region, China

• Published in: Building and environment Vol. 216; p. 108998
• Main Authors: Liu, Yiqiao, Chong, Wen Tong, Cao, Yijuan, Liu, Hongwei, Yu, Haowei, Cui, Tong, Chang, Li, Pan, Song
• Format: Journal Article
• Language: English
• Published: Oxford Elsevier Ltd 15.05.2022; Elsevier BV
• Subjects: Average daily window-opening probability; Binary logistic regression; Cold; Cold regions; Decision tree; Decision trees; Empirical analysis; Households; Indoor environments; Model accuracy; Modelling; Random forest; Regression analysis; Relative humidity; Statistical analysis; Summer; Typical window opening behavior; Winter; Random forest; Typical window opening behavior; Binary logistic regression; Decision tree; Average daily window-opening probability
• ISSN: 0360-1323; 1873-684X
• DOI: 10.1016/j.buildenv.2022.108998

Opening windows is an important factor when creating a comfortable indoor environment. However, research on window opening behavior in the Chinese region of hot summers and cold winters is limited. In addition, almost all previous research in this area has focused on the average characteristics of the sample mostly by building models of averageness, which largely erases individual behavioral attributes. To fill the gap, an empirical measurement and modeling study of window opening behavior was conducted in seven households in Zigong, Sichuan Province. The following was found: (1) The window opening behavior of residents in different cities in hot summer and cold winter regions significantly differed. The average daily duration of window opening by the test subjects in this paper is 1122 min/day, which is 2–5.5 times higher than previously published literature on this climate zone suggests. (2) Among the test subjects, three typical window opening behaviors were noticed based on the average daily window-opening probability (R), i.e., positive (R>95%), negative (R ≤ 5%), and high-intensity window opening (65% < R ≤ 95%). The first two categories refer to personal habits of residents independent of environmental and temporal factors. (3) High-intensity window opening behavior provided imbalanced data which is more accurately modeled by the random forest model (98.9% prediction accuracy) than the binary logistic regression and decision tree models, i.e., 14.5% more accurate than the former and 12.5% more accurate than the latter. Moreover, it was found that the relative humidity indoors is the factor that contributed the most to the accuracy of the model. •Significant differences in occupants' window-opening behaviors in the same climate zone.•Data samples regarding window-opening behavior should be individually analyzed.•Three types of behavior were identified based on "average daily window-opening probability".•Random forest algorithm had better prediction compared to logistic regression and decision tree.