Modelling Sewerage Systems with an Artificial Neural Network

Abstract

A new approach to non-parametric modelling of simulated flows in an urban drainage system is presented. The new model uses the storage of the drainage system as an input variable, in addition to the rainfall input, to a radial basis function neural network. Previous attempts to model this problem, using only rainfall inputs, have been unsuccessful because of the highly non-linear relationship between a rainfall event and the one dimensional flow associated with it (Figure 2). Compared with the traditional parametric model, the NN reduces greatly the time taken to produce simulated flows, and is much less expensive to set up. Furthermore, with respect to traditional methods, the NN predicts flows without a significant loss in accuracy. A practical use of the method is highlighted which aids hydraulic engineers in the making of environmentally sensitive decisions.