Water Distribution Networks Model Identification using Artificial Neural Networks

This paper proposes a novel scheme for model identification of water distribution networks (WDNs) based on artificial neural networks (ANNs). Generally, a model consisting of a set of the non-linear equations is iteratively solved to determine the hydraulic parameters of the WDNs and the process is computationally cumbersome. In this work, ANNs are used to enable the determination of hydraulic parameters at less complex computations. The dataset for the training of the ANNs is generated by means of hundreds of hydraulic simulations based on the pressure driven model. The dataset is then sorted into two categories implying that 2 different ANNs structure has to be defined. The demands and the pressure heads at the nodes are used as the inputs and outputs respectively to the first ANN model. In the second, the flows in the co-tree branches of the graph are used the output to the ANN structure. The results in this work show that both types of defined ANNs may be used to identify the model of the water distribution network. However, the first ANN model has a better estimation capacity at high training requirement.