Water Demand Analysis In Urban Region By Neural Network Models

Statistical water demand models are usually developed as time series coefficients using historically available water demand data, together with any other relevant variables. But structure identification turns out difficult for most of the applications. This study would count on the artificial neural networks (ANN) to forecast the water demand patterns. The ANN model may exhibit a nonlinear feature learned from historical data, in the same way as humans learn from experience. The nonlinearity, high complexity, and uncertainty associated with water demands may favor the potential use of ANNs to compete with or outperform the conventional time series methods for forecasting the similar topics. If the ANNs model is learned correctly, as verified by the accuracy of the predictions using input data not used during the training, the ANN algorithm can be robust with low computation time requirements, even if there are some errors or noise in the input data. Two types of cities, including Oviedo (fast growth) and Winter Springs (slow growth) in the Great Orlando Metropolitan Area in Florida were investigated with respect to monthly data. Such pattern recognition practices would help water utilities identify the expansion and operation strategies in water distribution systems in the long run. Copyright ASCE 2006