Optimal Meter Placements Based on Multiple Data-Driven Statistical Methods for Effective Pipe Burst Detection in Water Distribution System

Pipe burst events in the water distribution system (WDS) should be rapidly detected to secure the system functionality and reliability. Effective and efficient pipe burst detection (PBD) techniques can reduce water losses and prevent potential secondary failure (e.g., sinkhole) caused by ground liquefaction. This paper proposed an optimal meter placement method for PBD based on six data-driven statistical process control (SPC) methods. Total five meter placements are deduced by the optimization process, harmony search (HS) algorithm. They consist of two heterogeneous meters measuring different hydraulic quality characteristic values: pressure and flow rate. The optimal meter sets sought were compared with respect to two detection effectiveness indicators: true detection probability and false detection probability. The final optimal meter set should be generally effective for all utilized methodologies, therefore, the objective function is generated with those indicators and weights of them to reflect their effectiveness fairly. Three sets of synthetic time series data were produced from a network hydraulic EPANET model of Austin in Texas to demonstrate the proposed method: historical, normal, and abnormal data.