Leaks' detection in water distribution networks with demand patterns

A novel approach for the continuous detection of leaks in water distribution networks (WDNs) assuming uncertain demand patterns is presented. The proposal is based on a demand pattern construction with repetitive multiple statistic models which allow detection through residual generation for each corresponding model. The key of the method is the time division of the pattern such that the demand satisfies the stationarity condition during each time interval. Thus, the sequential set of demand models allows the use of traditional multivariate statistical tool such as principal component analysis (PCA) for the monitoring of each interval in a sequential manner. Simulations with an academic network are used to test the performance of the proposal, and the results show a lower false alarm rate than the standard PCA.