Comparative Analysis Of Window Opening Behavior Model Of Residential Buildings In Xi’an

Natural ventilation is the simplest and most energy-efficient way of ventilation.Opening windows can meet the needs of indoor personnel for fresh air,and maintain good indoor thermal comfort while improving indoor air quality and ensuring that the indoor pollutant concentration is not excessive.The ultimate purpose is to reduce the energy consumption of buildings.Therefore,the occupant’s window opening behavior is an important kind of human behavior in the building,and the accuracy of the window opening model has an important impact on the building simulation.Taking seven urban households in Xi’an as an example,the window opening behavior and indoor and outdoor air quality of the residential buildings were studied.The window states,indoor and outdoor environment,and air quality parameters were continuously monitored in seven households,including four households from August 1,2018 to August 1,2020,and the other three households from May to August 2020.Firstly,the monitoring data was preprocessed which the indoor and outdoor environment and air quality and window states data were integrated into the same time axis for analysis,and the data of two years were divided according to the standard of seasonal division in meteorology.Secondly,the multivariate analysis of variance and the Logistic regression equation were used to ascertain the major factors that affected the window opening behavior,and the correlation analysis of window opening behavior was conducted from the aspects of time,different seasons,different room functions and different factor levels.Furthermore,judging whether the raw data and prediction data were balanced.The method of redividing the training set was adopted for unbalanced data to establish the relevant models of the window opening behavior,including Logistic regression,BP neural network and Elman neural network.Finally,the accuracy and AUC value(Area Under ROC Curve)of the models were used to evaluate the model.The results show that the period of maximum window opening frequency in the living room and master bedroom of different residents are similar,and the window opening time of different room functions in the same households are also almost the same,and it is found that the window opening duration is the longest in summer.Considering the relationship that indoor and outdoor environmental parameters and air quality and the probability of window opening behavior,we can find that in a certain range,the window opening probability is positively correlated with indoor temperature and indoor PM2.5 concentration,while negatively correlated with the outdoor PM2.5 concentration.Moreover,the probability of window opening has a critical value with the change of indoor humidity,and the critical value of indoor humidity is 50%.The problem of sample imbalance can be improved effectively by redividing the training set for the raw data,thereby improving the AUC value of the model.According to the accuracy and AUC value of three models established in different seasons of 8 rooms,it was found in most of the cases that the Elman neural network is the best,followed by the BP neural network,and the Logistic regression model is the worst.On the whole,the neural network model is superior to Logistic regression.The study provides a certain theoretical basis for establishing the residential buildings model of window opening behavior in Xi’an.