Online burst detection in water distribution networks based on dynamic shape similarity measure

Monitoring water demand is extremely helpful in the early detection of issues and malfunctions in water distribution networks. Therefore, distinguishing abnormal water meter readings is an important problem in the management process. Although much focus has been given to this problem in the last few years, many of the approaches still lack efficiency and scalability when adapted to (near) real-time settings. To tackle these issues, we propose the Online Curvature Matcher, a real-time algorithm for the detection of pipe burst events in flow readings of District Metering Areas (DMA), which requires only a few weeks of data clean of burst events. In the proposed methodology, libraries of segments of the time series are constructed and then used to find abnormal patterns in consumption. We also propose the incorporation of additional constraints to address the complexities of real-world data (such as the existence of irrigation events, seasonal changes, and discrepancies in behavior due to weekday). We perform widespread testing on data from 32 DMA. The results indicate an increase in detected bursts compared to the state-of-the-art, while a decrease in the number of generated false positive events was achieved. Additionally, it was found that six weeks of history is the minimum amount of data necessary to perform online burst detection, while Dynamic Time Warping (DTW) is a suitable distance metric to measure the similarity between different flow reading segments.