An Ensemble Neural Network Model to Forecast Drinking Water Consumption

A reliable short-term forecasting model is fundamental to managing a water distribution system properly. This study addresses the problem of the efficient development of a deep neural network model for short-term forecasting of water consumption in small-scale water supply systems. These aqueducts experience significant fluctuations in their consumption due to a small number of users, making them a challenging task. To deal with this issue, this study proposes a procedure to develop an ensemble neural network model. To reinforce the ensemble model to successfully deal with the weekly and yearly seasonality which affect these data, two different time-varying correction modules are proposed. To constitute the ensemble model, the simple recurrent neural network, the long short-term memory, the gated recurrent unit, and the feedforward architectures are analyzed in two case studies. The results show that the proposed ensemble model can achieve a robust and reliable prediction for all four of the architectures adopted. In addition, the results highlight that the proposed correction modules can significantly improve the predictions.