Bad data analysis with sparse sensors for leak localisation in water distribution networks

Traditional bad data detection and localisation, based on state estimation and residual analysis, produces misleading results, with high rates of false positives/negatives, in the case of strongly-correlated residuals arising from a low redundancy of sensors. By clustering the measurements according to the structure of the residuals covariance matrix, a method is proposed to extend bad data analysis to the localisation and estimation of anomalies at the coarser resolution of clusters rather than single measurements. The method is applied to the problem of water leak localisation and a realistic test-case, on the water distribution network of a major European City, is proposed.