Comparison Of Different Occupant Window Behavior Modeling Approaches In Office Building

Window operation is an important occupant behavior,and has significant impacts on building energy consumption,indoor thermal comfort and air quality.Recently,various stochastic and non-stochastic models have been proposed,aiming to describe occupant window behavior based on several influencing factors.However,most of the employed methods are Logistic regression and Markov chain techniques,and the application of machine learning to model occupants'window behavior is rarely investigated.The prediction accuracy of former two algorithms were lower than machine learning algorithms.However,the high prediction accuracy is based on the large database which is hard for current data measurement in China.Furthermore,researchers have adopted different indices to evaluate the performance of their models,and thus there is a lack of a horizontal comparison among these models.In addition,most published studies referring to occupants'window behavior have been carried out within European countries,where the influence of outdoor air quality is rarely considered.However,the air quality has influenced the indoor window opening behavior obviously in recent years in China,and the parameter should be analyze.Support Vector Machine?SVM?was investigated in this paper which is a totally new algorithm for window opening model in order to bridge the gap between the algorithm and database.Whats more,three more models were built based on Logistic Regression,Markov chain and BP network,so that deliver a more comprehensive evaluation of different models developed based on multiple algorithms,under the same indices and the same datasets.And the PM_2.5concentration was considered as a influencing factor and analyzed in this paper.Indoor and outdoor temperature,outdoor relative humidity,wind speed,wind direction,sunshine hours,PM_2.5concentration,office occupied state and window states of the office building in a college in Beijing during transition season were monitored.And four models mentioned above were built based on the data.During the analysis,it was found that SVM yield prediction models of office window states with higher accuracy and better interpretability of highly correlated factors as compared to the other three algorithms.All of these algorithms arrive their top prediction accuracy under six influencing factors.Among them,the highest accuracy of Logistic Regression model,Markov chain model,BP neural network model and SVM model is 53%,57.9%,79.5%and 83.6%,respectively.Besides,anotherimportant found is when BP network algorithm is used to built occupant window opening behavior model which independent variable is 6 or less than 6,the training data should not less than 2880,once the database less than the limit,the stability of the model will be destroyed significantly.On the other hand,the accuracy of all models are improved when PM_2.5concentration was analyzed in each model.Thus,PM_2.5concentration should be considered as an influencing parameter for China window opening behavior.And the writer sincerely hope that the proposed approaches provide a new and accurate method for engineers and building operators to better understand occupant window behaviors and their impacts on energy use in office buildings.