Urban water demand forecasting based on HP filter and fuzzy neural network

Abstract

Urban water demand is a complex function of socio-economic characteristics, climatic factors and public water policies and strategies. Therefore a combination model is developed based on the multivariate econometric approach which considers these parameters to forecast and manage the urban annual water demand. Firstly, the factors correlative with water demand are selected, then the trend and cyclical components of the factors are calculated by the Hodrick–Prescott (HP) filter method. The multiple linear regression method is applied to simulate the trend components and the fuzzy neural network is built based on the cyclical components, and then the two models are combined to forecast the urban annual water demand. In order to illuminate the model, it is used to forecast the annual water demand of Dalian against actual data records from 1980 to 2007. By comparing with the traditional methods, the preferable model accuracy demonstrates the effectiveness of the fuzzy neural network and multiple linear regression based on the HP filter in forecasting urban annual water demand. After model testing, the sensitivities of the influence factors in the model are analyzed. The results show the model is reliable and feasible, and it also helps to make predictions with less than 10% relative error.