The same dimension gray recurrence prediction, multiple nonlinear regression prediction, GRNN neural network prediction methods were done to get domestic water demand of Xi''an, and then prediction error of different methods was compared. The paper proposect the combination forecast method based on shapely value which equal to the information contribtive ability of predictive value calculated the shapley value and combination weights of different forecast methods, formed combination forcast model for water demand. The result showed that the combination method had a smoothing error curve and a smaller average error value. It has a certain prediction accuracy and is applicable to water demand in the short-term water demand forecast.