Efficient automated quality assessment: Dealing with faulty on-line water quality sensors

Issue title: Environmental Data Mining

Guest editors: Karina Gibert

Article type: Research Article

Authors: Alferes, Janelcy^{; *} | Vanrolleghem, Peter A.

Affiliations: modelEAU, Universite Laval, 1065, Avenue de la Medecine, Quebec, Canada QC G1V 0A6. E-mails: janelcy.alferes.1@ulaval.ca, peter.vanrolleghem@gci.ulaval.ca

Correspondence: [*] Corresponding author. E-mail: janelcy.alferes.1@ulaval.ca.

Abstract: Current environmental challenges for water resources include guaranteeing good ecological status of water bodies, promoting sustainable water use and protection of water resources. A key aspect in the achievement of these objectives is the application of a consistent and efficient monitoring strategy. Implementation of continuous water quality measurement systems is allowing to capture the dynamics in water systems for identification of critical events, cause-effect relationships and trends among others. Huge amounts of data are then being generated with uncertain quality. Water quality monitoring networks will only be useful in practice if careful quality assessment, of the data is carried out. With a practical vision, this paper presents a method for automatic data quality assessment extracting information from individual water quality time series from on-line sensors. Data mining techniques based on forecasting models are used to detect and remove unreliable data from the raw data sets. A posterior analysis is applied to remove noise and detect abnormal situations and potential sensor faults. The proposed tool has been successfully tested on water quality time series collected from different water and wastewater systems.

Keywords: Data quality assessment, forecasting techniques, on-line water systems monitoring, fault detection

DOI: 10.3233/AIC-160713

Journal: AI Communications, vol. 29, no. 6, pp. 701-709, 2016