A Neural Network Approach to Simplify Mathematical Models of Urban Water Distribution Networks

Abstract

Computer-based operational control strategies in water industry require effective and accurate mathematical models of water distribution networks. The model usually consists of a set of nonlinear differential equations. The need for model simplification arises from the fact, that the network model must be simulated many times during optimization and only a very limited subset of variables is usually utilized by a cost function. In the paper a neural network approach to simplification of mathematical models of water distribution networks is discussed. As a case study, a comparison of dynamic behaviour of selected variables calculated by the full mathematical model of a water distribution network and their approximation by a neural network has been presented.