Occupants' window opening behavior in office buildings: A review of influencing factors, modeling approaches and model verification

Window opening behavior is a critical factor that affects the Indoor Environment Quality and energy consumption of buildings. However, inaccurate consideration of this behavior in performance simulation tools can lead to uncertainty between predicted and actual building performance. Several research studies have been conducted to understand the stimuli that influence occupant interactions with windows and develop predictive models. This study systematically reviews the literature related to window-opening behavior, aiming to analyze the influencing parameters, modeling approaches, and main findings. Out of 447 studies screened, 74 were selected for preliminary evaluation. For the final assessment, only studies conducted in office buildings were considered. While more than 20 factors have been identified to influence window-opening behavior, there is still a lack of consensus on the key triggers. Additionally, some climates and contexts have received little research attention, and some parameters have been underexplored, indicating the need for further studies. Modeling approaches were categorized into two groups: predicting window state using Bernoulli models and predicting occupant actions using discrete-time Markov chains. Recently, other machine learning methods, such as artificial neural networks, have been used as well. Finally, the reliability of the literature models was assessed by applying validation criteria to an external dataset. The validation results showed significant variation regarding the ability of the literature models to predict window-opening behavior. The results indicated that Models based on discrete-time Markov chains outperformed Bernoulli models.