A coupled Decision Trees Bayesian approach for water distribution systems event detection

Detecting contamination events in water supply systems is a constant concern for utilities. It is reasonable to assume that injection of foreign substances will affect the behaviour of typically measured water parameters. For this reason, identifying contaminants using water quality and hydraulic measurements which are regularly monitored is appealing. A generic framework integrating Decision Trees (DTs) and Bayesian sequential probability updating rule is presented for detecting contamination events in Water Distribution Systems (WDS). The Aquatic Event Detection Algorithm (AEDA) utilizes DTs to depict the correlation between water quality and hydraulic parameters in order to detect possible outliers. The analysis is followed by updating the probability of a contamination event by recursively applying Bayes rule. AEDA is assessed through correlation coefficient (R2), Mean Squared Error (MSE), confusion matrices, Receiver Operating Characteristic (ROC) curves, and True and False Positive Rates (TPR and FPR). AEDA is tested using simulated contamination events, imposed on water parameters, to imitate pollution scenarios in WDS.