Application of Neural Networks in Analysing Failures ofWater Distribution Systems

摘要

Although failures in water distribution system are inevitable, compromising the levels ofrnservice to the consumer during failures is not acceptable. It is common for a failure in therndistribution system to cause a reduction in pressures resulting in a reduced nodal flow tornconsumers. In order to predict the reduction in the levels of service as a result of thernreduced flows, it is important to relate pressure changes with nodal outflows duringrnfailure events.rnConventional network analysis methods do not allow the nodal outflow to be adjusted asrna result of pressure reduction, as the models in general are demand driven. Modifiedrnnetwork analysis is required where pressure dependent outflow functions are used.rnHowever many limitations associated with pressure dependent demand functions havernbeen reported in the literature and these limitations restrict their applicability to realrnnetworks.rnIn this paper a new method based on artificial neural networks is presented for relatingrnpressures and nodal outflows. The proposed method involves the detailed analysis ofrnseveral typical secondary networks, to provide sufficient data to train the neural network.rnOnce trained the neural network is used to calculate the reduced outflows to nodes, basedrnon the secondary network type, time of failure and the resulting reduced pressure. Thernpaper presents an example application to demonstrate the method and its strength overrnprevious methods.