Pressure sampling design in water distribution systems using graph partitioning and entropy

ABSTRACT Hydraulic models of water distribution systems (WDSs) play a crucial role in system planning, design, and management. The nodal water demand, a key parameter in such models, must be accurately estimated based on pressure measurements within the network. An appropriate pressure sampling design is essential for improving estimation accuracy. While existing studies have emphasized the spatial uniformity of monitoring points and the informational content of sampling schemes, they often fall short in terms of estimation accuracy and computational efficiency. This study proposes a novel sampling design method aimed at improving the accuracy and efficiency of nodal demand estimation in WDSs. The method first ensures a spatially uniform distribution of monitoring points through graph partitioning, and then selects the most informative nodes within each subgraph based on the entropy metric. When applied to a real-world network, the proposed method reduces the average absolute pressure deviation by up to 0.76 meters compared to existing approaches, while achieving a computational cost that is only one ten-thousandth of that required by traditional methods. Furthermore, the method demonstrates strong robustness under varying demand distributions.