Forecasting Daily Urban Water Demand Using Artificial Neural Networks, A Case Study of Tehran Urban Water

Water demand forecasting is one of the most important concerns for managers of water supply systems as the results can affect many decisions. Daily demand forecasting cannot be usually accomplished by mathematical functions because it is a complicated function of many variables. In this paper, neural networks are used to predict Tehran daily water demand. At first, weather data from three Tehran weather stations are weighted via the Thissen method and the effective input data parameters are selected using the regression of the weighted effective weather and consumption data. The effective parameters include daily average temperature, relative humidity, and last day to last week (7 days) as well as last year water consumptions. Three different ANN models are built in this stage: a three-layer model with one hidden layer including seven neurons, a four-layer model with two hidden layers including seven neurons in the first and four neurons in the second hidden layer, and a RBF three-layer model with twenty neurons in the middle layer. Comparison of the results of ANN with neuro-fuzzy and time series models shows that ANN models have a higher capability for predicting Tehran daily water consumption. Among these models, the ANN perceptron 3-layer model with a nonlinear output produced more accurate results