Neural networks for water demand time series forecasting

Abstract

One of the main problems in the management of large water supply and distribution systems is the forecasting of the daily demand in order to schedule the pumping effort and minimize the costs. This work presents the use of neural network models, with and without intervention series, to this forecasting task. The networks used have units with a variable gain in the transfer function and the training phase is performed like a standard optimization problem, in order to do it more efficiently.