A real-time anomaly detection algorithm/or water quality data using dual time-moving windows

Anomaly data in real-time water quality monitoring systems can cause false alarms and significantly decrease system stability and reliability. We propose a novel anomaly detection algorithm for water quality data using dual time-moving windows, which can identify anomaly data from historical patterns in real-time. The algorithm is based on an autoregressive linear combination model, a prediction interval with dual time-moving windows and a backtracking verification strategy. We have tested the algorithm using 3-month water quality data of PH from a real water quality monitoring station in a river system. Experimental results show that our novel anomaly detection algorithm can significantly decrease the rate of false positive and has better anomaly detection performance than AD and ADAM algorithms.