A robust localized soft sensor for particulate matter modeling in Seoul metro systems
Introduction
Metro systems have been considered to be an important and popular mode of transportation because they relieve congestion and provide an efficient solution to the problem of insufficient public transportation. Given the fact that more than eight million commuters utilize the Seoul metropolitan subway system daily, the indoor air quality (IAQ) in metro systems, particularly in underground subway stations, is very important to public health and has attracted a great deal of public attention [1]. In the case of the Seoul metropolitan subway system, the Korean Ministry of Environment has established regulations in order to monitor and control the levels of several hazardous pollutants, such as carbon monoxide (CO), carbon dioxide (CO2), nitrogen dioxide (NO2), particulate matters with diameters less than 10 (PM10) and 2.5 μm (PM2.5), airborne fungi, and carbonyl compounds. Excessive exposure to these hazardous pollutants, among which particulate matter is thought to be more toxic than the others, can cause serious health consequences, such as cardio-respiratory illnesses and mortality [1], [2], [3], [4], [5], [6]. In particular, the concentration of particulate matter, including PM10 and PM2.5, in underground subway stations is higher than in outdoor spaces in Seoul [7]. Therefore, developing accurate prediction models of particulate matter for monitoring the IAQ in subway stations has become an important issue of public concern.
Several prediction and monitoring techniques have been conducted in order to ensure passengers’ health in metro systems [8], [9], [10], [11]. Kim et al. [8] compared the prediction performances of PM10 and PM2.5 in a subway station using three data-driven prediction models, in which the recurrent neural networks method was more appropriate than the multiple linear regression and neural networks methods. Kim et al. [9] developed a predictive fault diagnosis system based on a multi-way principal component analysis in order to monitor the periodic patterns of the air pollutants in a subway system. Kim et al. [10] proposed season dependent models for the monitoring and prediction of PM10 and PM2.5 in a subway station. The season dependent models utilized the partial least squares (PLS) modeling method for prediction purposes. In the work of Liu et al. [11], a recursive partial least squares (RPLS) method, which is widely used to model dynamic processes, was used in order to improve the prediction performance of the indoor air quality soft sensors in an underground subway station.
Soft sensors are a key technique used to estimate reliable product quality or other important variables when online analyzers are not available or encounter failures. These sensors have been widely used in many process industries [12]. Taking subway systems as an example, IAQ sensors sometimes exhibit poor quality and low reliability due to the long term usage and the hostile underground environment. Therefore, some sensors may experience failures, including bias, drifting, complete failure, and precision degradation [13]. In this case, soft sensors are good alternatives to the faulty real sensors. According to Kadlec et al. [14], data-driven soft sensors can be categorized into statistical method-based and soft computing-based sensors. In the last few decades, linear statistical modeling methods, such as PLS and principle component regression (PCR), have been extensively used for soft sensors. On the other hand, nonlinear soft sensors, which can capture the nonlinear characteristics of the process variables, have already been intensively researched; examples of such soft sensors include the artificial neural network (ANN) [15], ANN-based PLS or PCR, support vector machine (SVM) [16], and adaptive neuro-fuzzy inference system (ANFIS) [17]. Derived from the SVM theory, the least squares support vector regression (LSSVR) is a new nonlinear prediction method. LSSVR has been applied to the modeling of many complex industrial processes including the Tennessee Eastman (TE) process [18] and the debutanizer column [19].
Many modified and hybrid methods, such as the kernel, dynamic, and adaptive extensions of the original soft sensors, have been developed as soft sensors [14]. For example, different recursive methods, such as the recursive support vector regression (RSVR) and recursive partial least squares (RPLS) [20], have been applied to match the time varying features of numerous processes. However, these adaptive methods suffer from the problems of blind and excessive updates when the process is running in a certain operation region. In addition, recursive methods cannot deal with abrupt changes in the process variables. In order to alleviate these problems, the just-in-time (JIT) learning technique, which has been applied in nonlinear process monitoring and soft sensors [21], [22], [23], [24], [25], [26], has been developed to cope with system dynamics as well as nonlinearity. In comparison to the traditional global modeling methods, the JIT learning technique exhibits the main characteristic of approximating a nonlinear system using several simple local models, which are only valid in a certain operating region. Therefore, the JIT learning technique is a local modeling method that can be applied in processes exhibiting highly nonlinear and dynamic characteristics. In the present study, two local models, including a linear PLS method and a nonlinear LSSVR method, were introduced into the JIT learning framework for the prediction performance improvement of the PM2.5 concentrations in a subway station.
Section snippets
IAQ data in Seoul metro systems
The air pollutants data used in this study were collected from a mini-volume air sampler and a telemonitoring system (TMS) installed at the center of a subway platform in Seoul (Fig. 1). The measuring principles of seven air pollutants are summarized as follows: (1) nitrogen monoxide (NO), nitrogen oxides (NO2), and NOx were measured by the chemiluminescence of the nitro-oxides materials; (2) PM10 and PM2.5 were measured using the beta-ray attenuation principle with the corresponding size
Monitoring system
The IAQ data in the underground subway stations exhibited periodic and nonlinear characteristics. Fig. 3shows the hourly average values of the number of passengers and the concentrations of CO2 and PM10 in four subway stations. The diurnal variations of the IAQ pollutants in one week are also shown in Fig. 3. A similar pattern existed between the number of passengers and the concentrations of CO2 and PM10. In terms of the CO2 concentrations, two peaks could be observed at approximately 10:00
Conclusions
The JIT-based methods using a local model structure could effectively track the time-varying changes of the process, which made it attractive for modeling nonlinear processes. Equipped with the nonlinear local modeling method of the LSSVR, a robust JIT-based soft sensor integrated with an outlier detection step was proposed for the development of a PM2.5 prediction model in an underground subway system. The prediction performance of the JIT–LSSVR was greatly improved; the RMSE and R2 of the
Acknowledgments
This study was supported by the Foundation of Nanjing Forestry University (No. 163105996) and the National Research Foundation of Korea (NRF) grant funded by the Korean government (MSIP) (No. 2015R1A2A2A11001120).
References (34)
- et al.
Characteristics of PM10, PM2.5, CO2 and CO monitored in interiors and platforms of subway train in Seoul, Korea
Environ. Int.
(2008) - et al.
Health risks of NO2, SPM and SO2 in Delhi (India)
Atmos. Environ.
(2005) - et al.
Temporal variation of airborne fungi concentrations and related factors in subway stations in Seoul, Korea
Int. J. Hyg. Environ. Health
(2006) - et al.
Characteristics and personal exposures of carbonyl compounds in the subway stations and in-subway trains of Shanghai, China
J. Hazard. Mater.
(2010) - et al.
A multivariate study for characterizing particulate matter (PM10, PM2.5, and PM1) in Seoul metropolitan subway stations, Korea
J. Hazard. Mater.
(2015) - et al.
Spatial distribution of particulate matter (PM10 and PM2. 5) in Seoul Metropolitan Subway stations
J. Hazard. Mater.
(2008) - et al.
Predictive monitoring and diagnosis of periodic air pollution in a subway station
J. Hazard. Mater.
(2010) - et al.
Monitoring and prediction of indoor air quality (IAQ) in subway or metro systems using season dependent models
Energy Build.
(2012) - et al.
Evaluation of passenger health risk assessment of sustainable indoor air quality monitoring in metro systems based on a non-Gaussian dynamic sensor validation method
J. Hazard Mater.
(2014) - et al.
Data-driven soft sensors in the process industry
Comput. Chem. Eng.
(2009)
A comparative study of just-in-time-learning based methods for online soft sensor modeling
Chemom. Intell. Lab. Syst.
A new data-based methodology for nonlinear process modeling
Chem. Eng. Sci.
Nonlinear process monitoring using JITL-PCA
Chemom. Intell. Lab. Syst.
Online monitoring of nonlinear multiple mode processes based on adaptive local model approach
Control Eng. Practice
Levels of particulate air pollution, its elemental composition, determinants and health effects in metro systems
Atmos. Environ.
Investigation of indoor environmental quality in Shanghai metro stations, China
Environ. Monit. Assess.
Data-driven prediction model of indoor air quality in an underground space
Korean J. Chem. Eng.
Cited by (39)
Soft sensor for predicting indoor PM<inf>2.5</inf> concentration in subway with adaptive boosting deep learning model
2024, Journal of Hazardous MaterialsAn overview of artificial intelligence in subway indoor air quality prediction and control
2023, Process Safety and Environmental ProtectionJust-in-time based soft sensors for process industries: A status report and recommendations
2023, Journal of Process ControlA novel deep learning model integrating CNN and GRU to predict particulate matter concentrations
2023, Process Safety and Environmental ProtectionA review on characteristics and mitigation strategies of indoor air quality in underground subway stations
2023, Science of the Total Environment