Analysis of uncertainties in water systems using neural networks

There are two main sources of uncertainty and errors in water distribution system simulations. The first one is associated with the modelling of physical elements and represents static (or slowly changing) inaccuracy of the network model. The second source of uncertainty has a dynamic nature and represents inaccurate predictions of consumptions and measurement noise. While, for instance, the pipes’ C factors change gradually over years or decades, consumptions and flows in the network change from minute to minute and are not well characterised as statistical processes (the difficulty of deriving and validating a probability density function for water use at some location at some particular moment).