Sustainable management of water distribution networks requires the timely detection of water leakages from pipelines. This will reduce wastage of resource, decrease cost of treatment and pumping, cut third party damage and reduce green house gas emissions. Some recently developed methodologies permit real time detection of pipe burst events by analyzing signals from pressure and flow meters located in District Metered Areas. These procedures are conceptually based on: (i) data preparation (e.g. de-noising; reconstruction); (ii) predictions based on data-driven models; (iii) identification of anomalies in flow/pressure and raising alerts based on a mismatch between model predictions and signals from meters. The paper analyzes the potential of the Evolutionary Polynomial Regression modeling paradigm in this framework. The idea is to use the Multi-Case EPR Strategy to develop flow/pressure prediction models using values recorded over a number of past time windows that are treated as separate data-sets. This means to have the same mathematical structure of the prediction model although with different sets of parameters, each minimizing the error over a different past time window. This, in turn, results in a range of predictions, each obtained using a different set of parameters, to be used for detecting anomalies and raising alarms. This approach is expected to have the following strengths: (i) the number of past time steps to be used for prediction is selected automatically by EPR from a set provided by the analyst; (ii) the EPR multi-objective paradigm returns a set of models which can be compared in terms of both selected variables (i.e., past time steps) and error statistics, thus avoiding over-fitting to past data; (iii) the range of predictions reflect different past time windows based on prior knowledge of the WDN in terms of demand/pressure patters, instead of purely probabilistic assumptions. The methodology is tested on a real network.
Berardi, Luigi; Laucelli, Daniele; Giustolisi, Orazio; and Savić, Dragan A., "Detecting Pipe Bursts In Water Distribution Networks Using EPR Modeling Paradigm" (2014). CUNY Academic Works.