A hybrid model for predicting window opening state in buildings based on non-intrusive monitoring

Abstract

Window opening behaviour is one of the most important factors for indoor air environment. The traditional models for window opening behaviour rarely focus on the window opening proportion, which has an important effect on optimal design of natural ventilation. A hybrid model combining the logistic regression model with a probability distribution model was proposed to analyse the window state distribution. A non-intrusive window monitoring method was used to sample window states, and the required sample size was analysed based on a pilot study. The Box-Cox data transformation was employed to establish a normal distribution model for the probability distribution of window opening state, and explore the relationship between outdoor temperature and the probability density function (PDF). The study found that the outdoor temperature, relative humidity and PM2.5 concentration had a significant effect on window opening states, and the outdoor temperature had a higher prediction accuracy (86.7%) for the logistic regression model. For different outdoor temperature, the parameters of PDF for window opening state were different. The mean and variance of the PDF were highest when the outdoor temperature was 20°C–25°C. This study can help to improve effective design and utilization of natural ventilation.