Real-time multivariate monitoring and diagnosis of air pollutants in a subway station

To control the quality of air pollutants, several key air pollutants are measured and monitored offline and also online by tele-monitoring (TMS) system which is built for the management of air quality. Until now, most of the monitoring methods has been used in an univariate approach which measures and monitors a single pollutant, such as particular matter(PM_2.5 or PM₁₀). Since it has a correlation relationship between variables, this study focuses on a multivariate statistical monitoring method for monitoring the indoor air quality in a subway. The proposed method consists of three main components: (1) principal component analysis (PCA) to reduce the dimensionality of multivariate air pollutant data and to remove collinearity; (2) global model to monitor nine real-time air pollution data and diagnose the status of indoor air quality during one year; (3) local models to keep track of the seasonal variations of air pollutants. The seasonal models are suggested to consider the variations of the pollutant concentration according to the climate change in Korea. The multivariate monitoring method is applied to a real time TMS dataset of nine air pollutants in a real subway station. It shows the accurate and reliable result of air pollutants in a subway over univariate monitoring, which can significantly enhance the power of the monitoring system. And the seasonal models allow to isolate the characteristics of the seasonal variations for specific monitoring of air pollutants. The multivariate approach is useful to check the indoor air quality status using all information of the sensors and to predict effects of indoor air pollutants to the passenger's health.