A hybrid approach based on benchmark simulation and the ensemble multivariate changepoint detection (EMCPD) is proposed to detect leakage occurrences. Bilinear bivariate spatial interpolation and Student’s t-test are further invocated to isolate leakage sites. First, a simulation was performed, assuming no leakage occurred to attain the benchmark working condition series of the L-town WDS. The pressure-dependent demand scheme was deployed with a simulation time step of five minutes. Upon simulation, the residuals of flowrates and pressures were calculated by contrasting the sensor measurements and the simulated results. The observed demand data were assimilated as the nodal pattern, and therefore the nodal demand residuals were omitted. Based on the flowrate residuals, the EMCPD was performed to detect the rough time of leak events. At this stage, the flowrate residuals series were averaged to three observations per day to denoise short term fluctuation, keep diurnal and weekly cycles, and accelerate the computation. Around each rough candidate, the EMCPDs were performed to distinguish accurate candidates. At this stage, both the flowrate residuals and pressure residuals series were engaged at five minutes level. Finally, around each accurate candidate, the nodal pressures were calculated by using bilinear spatial interpolation based on monitored pressure time series. The two-sample one-sided Student’s t-tests were performed to isolate the most likely sites. Since all pressures were estimated based on 34 sparse sensors, bilinear bivariate interpolation without extrapolation, as well as a p-value threshold of 0.01 were adopted, with the intention to avoid overfitting, to be conservative at un-monitored nodes, and to be parsimonious at raising alarms. The overall approach with limited assumptions on the underlying dataset distribution was both computationally efficient and conveniently transferable to other spatial-temporal datasets and mechanistic models. Future research could adopt spatial kriging to explore leakage isolation via water pressure interpolation among sparse sensors.

This publication is based upon work supported by the King Abdullah University of Science and Technology (KAUST) Office of Sponsored Research (OSR) under Award No: OSR-2019-CRG7-3800.