Abstract: The accurate prediction on urban water consumption is of great importance in management and improvement of water supply system.Traditional BP neural network prediction model is prone to problems like local minimum,weights adjustment and constant testament of parameters.Based on these confinements,genetic algorithms based on biological evolutionary theory was given to upgrade BP neural network algorithms,thus producing a new method namely GA-BP.Meanwhile,in view of the very low error accuracy resulted from unsuitable selection of input variables in traditional BP algorithms,this thesis analyzed the change rule of the urban water consumption to get suitable input variables.Then a prediction model was built with testament and simulation of historical data.The application of such prediction model was finally applied to a water supply company in Shenzhen.Its result indicates that the GA-BP is reliable and practical in predicting urban water consumption.

Key words: BP neural network,Genetic algorithms,GA-BP,City hourly water consumption prediction