Periodic pattern extraction and anomaly detection for free chlorine in drinking water network

Water contamination events are threatening the safety of drinking water. Free chlorine is widely used as the disinfectant in drinking water, which can be used as a surrogate parameter to provide indications of potential contaminants. In this article, the periodic fluctuation of free chlorine is studied and the fluctuation pattern is extracted by the singular vector decomposition method, and an anomaly detection scheme for free chlorine is proposed and tested. Firstly, the normal periodic pattern and current pattern of free chlorine are both extracted from the historical and online data, and then the difference between the current data pattern and the normal data pattern are compared with thresholds for anomaly declaration. The single point detection and data series detection are investigated for the purpose of short-term and long-term inspection. Further, the anomaly data treatment and the detection method using sub-patterns are discussed. Performance tests show that the proposed method is sensitive to the anomaly data, and is effective to detect anomalous condition in typical contamination scenes.