Study On Influencing Factors And Model For Occupant Window-opening Behavior: Case Study Of An Office Building In Xi’an

The indoor environment can be improved by opening windows.The behavior of opening windows can not only improve indoor thermal comfort and indoor air quality,but also reduce building energy consumption.At the same time,as an important input item of energy consumption simulation software,the window-opening behavior is significant to improve the accuracy of building energy consumption simulation.However,the behavior of window-opening is random,not only highly dependent on weather conditions,but also influenced by subjective will and personal habits,which makes it difficult to study the behavior of window-opening.Exploring the driving factors influencing window opening behavior,and establishing window-opening probability prediction models are significant to study the characteristics of window-opening behavior,which will be beneficial to the implementation of effective building energy-saving control technology in the future.Logistic regression model is the most widely used algorithm when establishing the window-opening behavior model,but its prediction accuracy is low,so it is necessary to explore more accurate models.There are few researches on using machine learning algorithm to simulate the window-opening behavior,but the introduction of machine learning can effectively improve the accuracy of window opening model.The research of machine learning algorithm in the field of window opening is still in its infancy.There are many types of buildings in China,and the behavior of window-opening is diverse in different types of buildings.This paper focuses on a typical office building in Xi’an as the research object.From September 15,2019 to August 31,2020,the window status,indoor and outdoor environmental parameters were measured.And the big office(larger than 90 m~2,larger than 15 persons),small office(the area of room is 32m~2,the person in room is 1～2)and conference room(with indoor staff only in use)were extensively analyzed.Secondly,based on the measured data,machine learning algorithms were introduced to establish window-opening behavior models for office buildings in Xi’an,which includes decision tree model,random forest model and Stacking fusion model,and the machine learning model and logistic regression used by traditional model were compared and analyzed.Finally,the simulation results of the model were compared with the measured data,and the important driving factors affecting the window-opening behavior of office buildings in Xi’an were obtained.The results show that the probability of opening windows is the highest when the staff first arrive at the office(72.1%in spring,65.1%in summer,63.0%in autumn,27.5%in winter),the probability of opening the windows decrease to a lower point during the lunch break(62.1%in spring,60.4%in summer,53.6%in autumn,and 23.1%in winter),and the probability of opening windows is the lowest.The average window opening time in the office is about 400minutes during the transition season and the cold season,and about 200 minutes during the heating season.The action of opening windows generally occurs when people arrive at the office,and the action of closing windows generally occurs when people leave the office.The most typical mode of window opening in summer is:open the window when arriving in the morning,and close the window when people leave.The results of window opening behavior models in this paper show that the introduction of machine learning can effectively improve the accuracy of the models.For a single learner,the accuracy of random forest model is the highest,ranging from 89.1%to 92.3%.Followed by the decision tree model,the prediction accuracy is between 84.3%～87.8%.The highest prediction accuracy of the logistic regression model is76.6%.The prediction accuracy based on Stacking fusion model is 0.2%～1.9%higher than that of random forest model.Based on the importance ranking of random forest features,indoor environmental parameters are more important than outdoor environmental parameters when analyzing the driving factors affecting the window state.By comparing the accuracy,complexity and the function of different models,the stochastic forest model is considered to be the best model to predict the window state.