Online Anomaly Detection in Multi-Parameter Sensor Data Stream for Water Networks

Online water quality monitoring is essential for water utilities to ensure public health and environmental safety. Water quality monitoring is crucial for anomaly detection and event classification, which is based on heterogeneous sensors that continuously generate multi-parameter time series data streams online. While many machine learning (ML)-based anomaly detection and event classification methods exist, most require historical batch data to learn patterns and are therefore not suitable for online, multi-parameter data streams. Therefore, this work presents an online anomaly detection framework based on unsupervised machine learning. The system ingests heterogeneous, multiparameter sensor data streams in real-time, performs data preprocessing, evaluates thresholds against established guidelines, and identifies potential anomalies using incremental Density-Based Spatial Clustering of Applications with Noise (DBSCAN), an unsupervised clustering approach. Potential examples of anomalies include pollution events, sensor fouling events, or runoff events. The analysis of error metrics shows that the proposed approach can identify, detect, and classify anomalies online with significant precision.