Water quality event detection and customer complaint clustering analysis in distribution systems

Abstract

Safe, clean drinking water is a foundation of society and water quality monitoring can contribute to ensuring this. A case study application of the CANARY software to historic data from a UK drinking water distribution system is described. Sensitivity studies explored appropriate choice of algorithmic parameter settings for a baseline site, performance was evaluated with artificial events and the system then transferred to all sites. Results are presented for analysis of nine water quality sensors measuring six parameters and deployed in three connected district meter areas (DMAs), fed from a single water source (service reservoir), for a 1 year period and evaluated using comprehensive water utility records with 86% of event clusters successfully correlated to causes (spatially limited to DMA level). False negatives, defined by temporal clusters of water quality complaints in the pilot area not corresponding to detections, were only approximately 25%. It was demonstrated that the software could be configured and applied retrospectively (with potential for future near real time application) to detect various water quality event types (with a wider remit than contamination alone) for further interpretation.