Abstract: At present, a massive portion of data stored in the heat-supply network management system has formed the data grave which does not embody the intrinsic properties of data. To solve this problem, it is particularly important to take effective mining methods which reuse the existing historical data to improve the current system. In this paper, we first dispersed the continuous attribute information based on both entropy and importance of attribute to preprocess the data of heat-supply network. Then we exploited three kinds of algorithm for data mining, namely, classification and prediction based on the decision tree, cluster analysis based on the K-mean partition and association rules mining based on the frequent itemset model. Finally, we established forecasting model combining the results of three aforementioned mining schemes. The model was then embedded into the prediction module of present system and the results demonstrated the proposed scheme can improve the prediction performance efficiently.