Model-Based Real-Time Detection of Contamination Events

We provide a preliminary set of results to evaluate a model-based algorithm for detecting anomalous distribution system water quality events in real time. This model-based algorithm relies on a real-time hydraulic and water quality model to estimate the water quality signature, for comparison with sensed water quality signals. The difference between the modelled and the sensed water quality would be processed (e.g., filtered) to detect changes that are inconsistent with noise and thus likely to be associated with a contamination incident. Our preliminary results evaluate the damaging influence of errors in model predictions caused by errors in the estimation of real-time water demands using a predictor-corrector algorithm. For one test network, even when the chlorine predictions are corrupted by significant appears qualitatively different from the noise generated by demand errors, providing hope that such signals could led to improvements in both sensitivity and specificity for event detection.