Author(s): Mengke Zhao; Dengming Yan
Linked Author(s):
Keywords: Leakage zone identification; Multi-time step pressure data; Water supply systems; XGBoost
Abstract: Water supply systems (WSSs) are vital infrastructure for cities. However, leakage occurs frequently due to factors such as long service life and imperfect management. It not only compromises the level of service but also leads to significant water loss. This paper proposes a novel algorithm for leakage zone identification based on multi-time-step pressure data analysis. First, the WSSs are partitioned into several leakage zones using k-means clustering algorithm. Subsequently, pressure data collected over consecutive time steps are subjected to multi-time-step feature fusion. An XGBoost model is then employed as the base classifier, where the input features are the fused multi-time-step pressure data, and the output is the identified leakage zone label. Results show that the localization accuracy improves as the number of time steps used for feature fusion increases, rising from 0.804 (using single time step) to 0.905 (using 10 consecutive time steps). However, further increasing the number of time steps cannot yield a significant improvement in the accuracy. Furthermore, the performance gain from using multiple time steps is more pronounced for leakage events with smaller leakage flows. For incidents with larger leakage flows, the model's accuracy plateaued after integrating data from just a few time steps.
Year: 2026