Abstract:With the rapid development of smart grid, a great amount of related usage data in the smart grid is produced every day. Then how to make use of these data and mining the potential information become a huge challenge. Based on the analysis of user power data, by combining the KL- divergence method with the improved K-means algorithm, we give a new user feature extraction algorithm, which is used to describe the difference of power grid users. Finally, we validated the effectivity and efficiency of our algorithm by experiments.