Bentley Fellow Bentley Systems Inc., United States
Abstract Submission: A novel cloud-hosted software solution, Anomaly Leak Finder (ALF), has been developed in collaboration with PUB, Singapore's National Water Agency, to enhance leak detection and localization in 3 of Singapore's supply zones. These networks span over 1,000 km of underground pipelines and are monitored by approximately 95 smart sensors. ALF leverages near real-time hydraulic time-series data, combining data-driven prediction (DDP) and physics-based simulation (PBS) models to minimize non-revenue water (NRW) losses by detecting and localizing hidden leaks before they escalate into disruptive events. To further improve ALF’s near real-time (NRT) anomaly detection and localization accuracy, the minimum night flow (MNF) hour analysis methods, using MNF flow and pressure features, have been developed and integrated into ALF. ALF’s MNF technology detects small and subtle changes in flow and pressure behavior during MNF hours, using a moving average total pressure and net inflow analysis combined with linear regression to detect pressure downward and net inflow upward trends, which are indicative of potential hidden leaks in the network. This newly integrated analysis complements ALF's 24-hour outlier detection using DDP and PBS models based on statistical process control (SPC) methodology. The combined approach has been tested through hydrant flow tests and actual reported pipe leak scenarios in the field. This paper will thus present the complete results obtained from the above-mentioned tests using the MNF and 24-hour outlier detection analyses, followed by highlighting the ongoing improvements to ALF’s capabilities to address various notable challenges (e.g., extremely small leaks) in the field.
.jpg "Zheng Yi Wu photo")
