CCWI2017: F110 'A model pre-processing approach for improving calibration-based leakage detection using a genetic algorithm'

The paper presents a systematic approach for narrowing down the search for leaks and unknown closed valves in the water distribution network. The developed approach is applied on a real system and a calibration problem is solved for the ultimate purpose of detecting existing background leakage hotspots. A Genetic Algorithm is used to solve the optimization problem searching for calibration parameter values, while minimizing the differences between observations and model outputs. The optimisation problem is coded in two ways, a scenario-based framework where the maximum number of leaks and closed valves in the network is specified and non scenario-based framework. The leak detection methodology takes advantage of the new pre-processing method to reduce the search space size for the optimisation problems to only significant parameters that contribute to the fitness and hydraulic changes of the model. Artificial calibration data are generated by means of hydraulic modelling employed to mimic planned hydrant discharges during a low demand period. The staged approach demonstrates that the search for location and range of flows for unknown leaks can be reduced to only a small part of the network components. This appears to provide additional benefits towards calibrations problem complexity reduction and reduced time in finding leaks.