Detection of water quality failure events at treatment works using a hybrid two-stage method with CUSUM and random forest algorithms

Abstract Near-real-time event detection is crucial for water utilities to be able detect failure events in their treatment works (WTW) quickly and efficiently. This paper presents a new method an automated, near-real-time recognition of at WTWs by the application combined statistical process control machine-learning techniques. The resulting novel hybrid CUSUM system (HC-ERS) uses two distinct methodologies: one fault level individual quality signals second faulty processes WTW level. HC-ERS was tested validated on historical real-life UK WTW. methodology proved effective events, achieving high true-detection rate 82% with low false-alarm (average 0.3 false alarms per week), reaching peak F1 score 84% as measure accuracy. also demonstrated higher accuracy compared CANARY methodology. When applied real-world data, showed capability automatically reliably, hence potential practical industry.