The localisation of leaks and failures in water suply system (WDN) requires a reliable WDN model. For this purpose, calibration is carried out, i.e. the hydraulic simulation results are reconciled with the measurement results by adjusting the model parameters, primarily the pipeline roughness coefficients. The latest publications propose calibration algorithms extended with a neural network component (ANN), which aims to increase the reliability of the calibration result by estimating the reference pressure at all points of the pipe model based on measurement data.
In the course of the project, a case study was carried out to apply the calibration algorithm with the ANN component to a real WDN model and unique measurements obtained through research in the WaterPrime project. The algorithm was examined and proposals for its development were presented by modifying the existing ANN component and introducing pipeline grouping before the actual calibration stage.