Event Detection in Water Distribution Systems from Multivariate Water Quality Time Series

Abstract

In this study, a general framework integrating a data-driven estimation model with sequential probability updating is suggested for detecting quality faults in water distribution systems from multivariate water quality time series. The method utilizes artificial neural networks (ANNs) for studying the interplay between multivariate water quality parameters and detecting possible outliers. The analysis is followed by updating the probability of an event, initially assumed rare, by recursively applying Bayes' rule. The model is assessed through correlation coefficient (R(2)), mean squared error (MSE), confusion matrices, receiver operating characteristic (ROC) curves, and true and false positive rates (TPR and FPR). The product of the suggested methodology consists of alarms indicating a possible contamination event based on single and multiple water quality parameters. The methodology was developed and tested on real data attained from a water utility.

Performance Evaluation for a Contamination Detection Method Using Multiple Water Quality Sensors in an Early Warning System

“…The second approach to event detection is based on signal processing and data driven techniques [10,[15][16][17][18][19][20]. For example, Hart et al [15] reported a linear prediction filter (LPF).…”

“…Allgeier et al [17] and Raciti et al [18] utilized artificial neural networks (ANN) and support vector machines (SVM) to classify water quality data into normal and anomalous classes after supervised learning training. Perelman et al [19] and Arad et al [20] reported a general framework that integrates a data-driven estimation model with sequential probability updating to detect quality faults in water distribution systems using multivariate water quality time series. A common feature of the methods mentioned above is that they are merely relying on data process.…”

Detection

“…Perelman et al (2012) offered a comprehensive solution to event detection that relies on artificial neural networks with application of Bayes's rule and assessment of the model occurring using correlation coefficients, mean squared error, confusion matrices, receiver operating characteristic (ROC) curves, and true and false positive rates. Perelman et al (2012) tested their approach on a real data set. Arad et al (2013) followed up on the work of Perelman et al (2012) by adding a dynamic threshold scheme.…”

“…Perelman et al (2012) tested their approach on a real data set. Arad et al (2013) followed up on the work of Perelman et al (2012) by adding a dynamic threshold scheme. The decision variables were: positive and negative filters, positive and negative dynamic thresholds and data window size.…”

Performance evaluation for three pollution detection methods using data from a real contamination accident

“…Allgeier et al (2005) and Raciti et al (2012) utilized artificial neural networks (ANN) and support vector machines (SVM) to classify water quality data into normal and anomalous classes after supervised learning. Perelman et al (2012) and Arad et al (2013) reported a general framework that integrates a data-driven estimation model with sequential probability updating to detect quality faults in water distribution systems using multivariate water quality time series. In general, these algorithms process the water quality data at each time step and compare this data with a preset threshold.…”