Abstract: Aiming at the insufficient mining of peak load characteristics by user power consumption behavior analysis under the background of current electric power big data, an analysis method considering peak load characteristic indicators is proposed. Firstly, the definition of peak load and peak load characteristic index are explained, and the peak characteristic feature set is constructed according to peak load characteristic index. Then, K-means algorithm is used to cluster the feature set and obtain clustering result labels, and evaluate different clustering performances of different categories by contour coefficient. Finally, power consumption characteristics of different label users are analyzed. Simulation calculation is carried out by using open-source user electricity data of National Renewable Energy Laboratory. The results show that the use of peak characteristic feature set has a better clustering effect than the original user data set.