Modeling and analyzing patterns of residential manual window operation

Window operating behaviour can improve indoor air quality, human thermal comfort, and building energy efficiency. Studies on occupants’ window opening behaviour in hot summer and warm winter region of China are limited and influencing factors and prediction models are not clear. Another limitation is the large number of proposed machine learning-based window opening behaviour models. However, the applicability and stability of these models in different datasets has not been proven. In response to these questions, modelling and field measurements were conducted in Quanzhou, China. Two different types of window-opening behaviour were noticed in the tested households. The first type was the all-closed windows, which had an average daily window-opening rate of 0.03%. The second type was the low-intensity window opening. The average daily window-opening rate was 10.6% and 9.1%, respectively. Then, the analysis of point biserial correlation coefficients revealed different reasons for closing windows in low-intensity households. One household closed the windows due to high outdoor humidity and the other mainly due to high outdoor wind speed and outdoor temperature. Furthermore, the suitable hyperparameters were screened for the support vector machine (SVM) model by K-fold cross-validation and grid search. The prediction model achieved an accuracy of 98.5% on the test set. Finally, the SVM model was trained and tested to verify the robustness of the model using data from the published literature. The prediction accuracy was improved from 0.7% to 7.4% compared to the different models used in the published literature.