Reasonable prediction of future water demand is of great significance to realize the rational planning and scheduling of water resources. In order to predict the short-term water consumption accurately in Jilin province, the grey correlation degree-set pair analysis classified prediction model(GCD-SPACPM) was set up. And the accuracy of the model prediction was cross-checked by the actual water consumption of Jilin Province. Results showed that the relative error of the prediction data to actual data in 2015 was 2.00%, indicating a better prediction than Grey Prediction model and BP neural network. The mean error of data inspection of two decades was 2.675%, the predicting effect was good and it can be used for the regional water consumption prediction. According to this model and the development plan, the amount of water consumption of Jilin Province will reach 13.84 billion cubic meters by 2020.