New statistical method to detect ozone pollution hot spots and monitor instrument failure; combining PCA and MEWMA
A new statistical method developed by researchers at KAUST (King Abdullah University of Science and Technology, Saudi Arabia) can detect abnormal ozone levels within large bodies of monitored data. The monitoring methods can quickly and accurately detect ozone anomalies—localized spikes in ozone concentration indicated by sensor data.
The method could be used as an automatic tool, and could act as an early warning system for dangerous pollution levels and potential technical problems, said Assistant Professor Ying Sun from the University’s Computer, Electrical and Mathematical Science and Engineering Division.
Ozone is the reactive form of oxygen that contains three atoms per molecule (O3) rather than the normal two. Ground-level ozone is created by chemical reactions between other pollutants, especially oxides of nitrogen and carbon-based compounds released in vehicle exhausts and by many industrial processes. The reactions that create ozone are promoted by bright sunlight, often leading to photochemical smog. Exposure to ozone can cause breathing difficulties, eye irritation and other health problems, and may also harm crops and other vegetation.
The complexity of ozone (O3) formation mechanisms in the troposphere makes the fast and accurate modeling of ozone very challenging. In the absence of a process model, principal component analysis (PCA) has been extensively used as a data-based monitoring technique for highly correlated process variables; however, conventional PCA-based detection indices often fail to detect small or moderate anomalies. In this work, we propose an innovative method for detecting small anomalies in highly correlated multivariate data. The developed method combines the multivariate exponentially weighted moving average (MEWMA) monitoring scheme with PCA modeling in order to enhance anomaly detection performance. Such a choice is mainly motivated by the greater ability of the MEWMA monitoring scheme to detect small changes in the process mean.—Harrou et al.
In the KAUST team’s approach, a MEWMA control scheme is applied on the ignored principal components of the PCA model (which have smallest variances) to detect the presence of anomalies.
To test the method in the field, the KAUST researchers collaborated with a French team with access to data from a network of air quality monitoring systems in Normandy.
The results in France confirmed that the PCA-based MEWMA anomaly detection scheme can offer improvements on existing methods, but further work is needed to reduce the level of errors that might spark false alarms.
Sun and her colleagues also hope to apply their data analysis techniques to dust pollution, another major environmental issue in Saudi Arabia.
Harrou, F., Kadri, F., Khadraoui, S. & Sun, Y. (2016) “Ozone measurements monitoring using data-based approach,” Process Safety and Environmental Protection 100, 0957–5820