The prediction of urban water demand can provide reference for the planning and management of water resources, and it is of great significance to alleviate the pressure of regional water resources. By comparing the prediction accuracy of random forest, support vector machine and BP neural network model, this paper seeks the most suitable model for forecasting water demand of Yangtze River Delta urban agglomeration. The results show that the average relative errors of SVM in the fitting and forecasting stages are 2.64% and 2.20% respectively, and the overall error is 2.57%. Compared with the other two models, SVM has the best overall performance and can be used to predict the water demand of the Yangtze River Delta urban agglomeration.