This paper proposes a functional recommendation service of the electric power system based on server logs mining. First of all, we get the user log data from server los, and then preprocess those log data with "dirty" data. Secondly, we calculate interest-measure of each user pairs by the processed data sets, and we divide data set of interest-measure of each user pairs into multiple classes with similar interest-measure based on improved K-means clustering algorithm. Finally, personalized functional recommendation method is provided to each user. The experimental results prove the effectiveness of our method in electric power business system.