The Research Of Urban Water Demand Prediction

Urban water demand is growing with the fast growing urban population and economic development. Due to the current water shortages and inadequate water supply facilities, the contradictions of urban water supply and demand have becoming increasingly prominent. Urban water demand prediction is the basis for the plan of water supply, water use and water conservation, therefore, urban water demand prediction has important and far reaching significance.Generalized regression neural network has strong ability of non-linear fitting, it has strong advantages in the approximation ability, classification ability and study speed,it’s suitable for the forecasting analysis. The only downside is that its adjustable parameters- smooth factor is difficult to determine, if the value of the smooth factor is inappropriate, it will affect network performance. In this thesis, the niche particle swarm optimization(NPSO) is used for optimizing the adjusting parameters of GRNN-smooth factor after the summary comparison analysis of PSO and genetic algorithms.Smooth factor is mapped to the particle, finding the global optimum smoothness factor by particle optimization, determining the input and output variables, then the city water demand prediction model is constructed based on NPSO-GRNN.Through investigation and analysis of Beijing’s economic development,meteorology and hydrology and the history of supply water information, the related affect factors of water consumption in Beijing are analyzed by person correlation analysis method and two-sided test method in SPSS software, finding the key factors influencing the water consumption. NPSO-GRNN urban water demand prediction model, BP neural network urban water demand prediction model and GM(1,1) urban water demand prediction model were constructed and used to predict. The results showed that: the prediction accuracy of water demand forecast model are feasible.However, the predict effect of NPSO-GRNN prediction model is not only better than other methods, but also has advantages of higher prediction accuracy, easy to fall into local minimum value and less adjustment parameters. it provides a new method for urban water demand prediction.