Water demand prediction is a complicated muhifactor, with the multi-level non-linear system influenced by urban population, industrial and social economic level. And water demand is the important reference basis in decision of regional industrial structure and arrangement form. As a centralized parameter forecast method, support vector machine may generalize the future sample performance. Application of this method may yield good simulation and prediction results to solve problems that lack of clear physical structure and enough data. So, this paper applied SVM method to build SVM water demand prediction model, and coupled SVM with AGA to optimize parameters of SVM model. The model was applied as a case in Zhuhai. The results show that compared with simple GA, AGA has a better efficiency in SVM model parameter optimization, and compared with BPNN model, SVM model can make a more precise forecasting by solving problems such as little sample, empiricism and etc.