A fault-injection-based approach to leak localization in water distribution networks using an ensemble model of Bayesian classifiers

Leak localization in water distribution networks (WDNs) is essential to water management systems. Developing a reliable and robust leak localization technique is crucial for reducing water losses in large-scale WDNs. The challenge of leak localization in WDNs can be addressed using a model-based data-driven approach. Each node of a WDN is used as a category label by the classifier model for identifying leakages. As a result of the small number of sensors in a network and the uncertainties associated with the model parameters, the characteristics of some classes may be similar. Consequently, the algorithm's accuracy decreases. The method proposed in this paper uses the concept of fault injection and propagation in digital systems to separate the leak characteristics of the classes. This separation is done by creating a second leak with a known amount and location to propagate the effect of the first leak on the sensors. Various data sets are generated for each time sample by considering different values for the second leak. Finally, an ensemble of Bayesian classifiers simultaneously processes this data set to determine the leak's location. The accuracy of the proposed method has been evaluated using Hanoi and Modena network benchmarks. Experimental results show that the proposed method has 94.6% and 93.76 accuracy without using a time horizon technique, in the presence of all uncertainties for the Hanoi and Modena networks respectively which show more accurate results than existing methods.