A probabilistic multi-variable hybrid approach to window operations and indoor comfort in residential dorms

Occupant window interaction is a critical component in optimizing energy consumption and indoor environmental quality (IEQ). Understanding the influence of environmental and behavioral factors on window state decisions remains a significant challenge in building management systems (BMS). We present a hybrid probabilistic model to assess thermal comfort and predict the probability of the occupant opening or closing the window. The data was acquired from an open-source platform that provided yearly university dormitory window interactions. Bayesian networks (BNs) and logistic regression (LR) models were applied to predict the window-opening behavior of the occupants. An average accuracy of 92% for Bayesian and 94% for LR were obtained. The results were further enhanced by combining these models through weighted methods, with weights extrapolated through generative recursive iterations generating an average accuracy of 95% and Area Under the Curve (AUC) of 98%. The proposed hybrid approach significantly improves over existing predictive models in thermal comfort and window state prediction.