Journal of Hazardous Materials 183 (2010) 448–459 Contents lists available at ScienceDirect Journal of Hazardous Materials journal homepage: www.elsevier.com/locate/jhazmat Predictive monitoring and diagnosis of periodic air pollution in a subway station YongSu Kim a , MinJung Kim a , JungJin Lim a , Jeong Tai Kim b,∗∗ , ChangKyoo Yoo a,∗ a Department of Environmental Science and Engineering, Center for Environmental Studies/Green Energy Center, Kyung Hee University, 1 Seochon-dong, Giheung-gu, Yongin-si, Gyeonggi-do 446-701, Republic of Korea b Department of Architectural Engineering, Center for Sustainable Healthy Buildings, Kyung Hee University, 1 Seochon-dong, Giheung-gu, Yongin-si, Gyeonggi-do 446-701, Republic of Korea article info Article history: Received 3 March 2010 Received in revised form 6 July 2010 Accepted 10 July 2010 Available online 16 July 2010 Keywords: Air quality monitoring Lifted model Multiway principal component analysis (MPCA) Periodic pattern Predictive monitoring abstract The purpose of this study was to develop a predictive monitoring and diagnosis system for the air pollu- tants in a subway system using a lifting technique with a multiway principal component analysis (MPCA) which monitors the periodic patterns of the air pollutants and diagnoses the sources of the contamination. The basic purpose of this lifting technique was to capture the multivariate and periodic characteristics of all of the indoor air samples collected during each day. These characteristics could then be used to improve the handling of strong periodic fluctuations in the air quality environment in subway systems and will allow important changes in the indoor air quality to be quickly detected. The predictive monitor- ing approach was applied to a real indoor air quality dataset collected by telemonitoring systems (TMS) that indicated some periodic variations in the air pollutants and multivariate relationships between the measured variables. Two monitoring models – global and seasonal – were developed to study climate change in Korea. The proposed predictive monitoring method using the lifted model resulted in fewer false alarms and missed faults due to non-stationary behavior than that were experienced with the con- ventional methods. This method could be used to identify the contributions of various pollution sources. © 2010 Elsevier B.V. All rights reserved. 1. Introduction Advanced monitoring and control strategies for atmospheric environments are attracting renewed interest due to increasingly stringent environmental regulations, because concerns about the effects of the air quality of indoor microenvironments on public health are increasing. The Korea Ministry of Environment (MOE) established the indoor air quality (IAQ) act in an attempt to control major pollutants including PM 10 , CO 2 , CO, VOC and formaldehyde in indoor environments such as subway platforms. There is a strong interest in monitoring air quality to quickly detect and identify any fault or abnormality that might negatively affect the air qual- ity, because of the increasingly stringent air quality requirements imposed by law [1]. Many variables in the systems process or environment are recorded on-line or off-line in modern systems, and the number of variables recorded is increasing due to the development of new electronic sensors. Proper techniques are required to extract useful ∗ Corresponding author. Tel.: +82 31 201 3824; fax: +82 31 202 8854. ∗∗ Corresponding author. Tel: +82 31 201 2539; fax: +82 31 206 2109. E-mail addresses: jtkim@khu.ac.kr (J.T. Kim), ckyoo@khu.ac.kr, ChangKyoo.Yoo@biomath.ugent.be (C. Yoo). information from the extensive amount of recorded data [2]. Sub- way station sites in a metro are considered to be air quality “hot spots,” which also include heavily trafficked roadways and power plants. Indoor air quality in subway stations can be strongly influ- enced by one pollution source, particularly if the station is located downwind of the source [3]. Therefore, a major objective of the monitoring system is to quickly detect the occurrence of assignable causes of contaminated air quality so that detailed measurements and steps to correct ventilation issues can be undertaken in order to improve the indoor air quality. The current status in a system should be monitored in order to meet all of the operational targets for quality, safety constraints and environmental constraints at a minimum cost. In metro systems, traditional monitoring systems have been based on time-series analysis which measures and monitors a sin- gle particulate matter pollutant (PM 10 and/or PM 2.5 ), which is also known as univariate monitoring. Univariate monitoring charts are widely used to monitor a small number of key pollutant variables in an air pollution monitoring system that is capable of detecting the occurrence of any event having a special or assignable cause. However, monitoring only a single or a few variables is not ade- quate, because many variables are correlated and interrelated and therefore have an effect on one another. More specifically, univari- ate diagnostics is not sufficient for detecting highly concentrated 0304-3894/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.jhazmat.2010.07.045