A comparison of model-based methods for leakage localization in water distribution systems

Model-based methods for leakage localization in water distribution systems have recently been gaining more attention. These methods identify the leakage position by comparing the measured network data with the corresponding values simulated by a hydraulic model. In this study two model-based methods already proposed in literature, one based on the Sensitivity Matrix method and the other one on the Linear Approximation method, are analysed and compared to each other. The methods are applied to the same case study network, exploiting only data provided by pressure sensors. Various analyses are undertaken in order to investigate the main critical issues tied to the two methods, i.e. a) the use of different amounts of data averaged over different time windows, b) the impact of the model’s accuracy in terms of water demands and pipe roughness, and c) the effect of the number of pressure measuring points. The results show that higher efficiency is obtained by considering the hourly averaged data all together. Moreover, the Linear Approximation method is on average 3 times more accurate than the Sensitivity Matrix when a perfect hydraulic model is used, even with a reduced number of pressure sensors. However, when a hydraulic model and/or measured data affected by errors are considered, the Sensitivity Matrix is more accurate, with an average error almost 10% lower than the Linear Approximation.