Near Real-Time Anomaly Event Localization by Pressure Drop Interpolation, Clustering, and Parallel Optimization of Hydraulic Model Calibration

Abstract

Localization of anomaly events in near real-time (NRT) in water distribution networks is compelling but challenging for water utilities. This paper presents an integrated approach using both data-driven and hydraulic model–based methods to localize NRT anomaly events. Upon detecting an NRT anomaly event, the pressure drops at sensor locations are calculated, followed by estimating the pressure drops at junction nodes via an inverse-distance weighted interpolation method. Clustering is then performed based on the junction pressure drops and network topology to segregate and reduce the search areas. A genetic algorithm optimization is then performed with hydraulic model simulations to further locate the anomaly hotspots. The efficiency of the anomaly localization is enhanced by message passing interface (MPI)–based parallelization of optimization to facilitate NRT operations. The integrated method has been tested on both simulated simultaneous multiple leaks and real leakage events with field data, where the ground-truth leaks have been successfully covered in the clustered search areas. Leak hotspots are further pinpointed by hydraulic model–based parallel optimization within a distance of 300 m to the ground-truth leaks, indicating satisfactory accuracy in leak localization.