Since the distribution of time-of-use electricity in the energy Internet not only provides a basis for the formulation of time-of-use electricity prices, but also represents users" awareness of electricity consumption, an improved K-Means++ clustering algorithm and a typical user screening model are proposed to mine time-of-use electricity. Firstly, K-Means++ is improved, using simulated annealing algorithm (SA) and median threshold segmentation to automatically determine the initial centroid and number of clusters. The weighted compound of Frecher and Euclidean distance is used as a measure of similarity, and the weight is determined by Information entropy and Analytic Hierarchy Process (AHP) determine. Then the time-sharing power is clustered, and typical users are screened according to the typical user screening model from the clustering results of each cluster, and 3 types of power consumption are obtained. Finally, the industry power consumption behavior is analyzed from the perspective of main power consumption and the transformation of the power consumption. Analysis and get the same electricity consumption behavior in different industries. It is helpful for the power supply side to initially control the electricity consumption characteristics of the regional industry electricity consumption groups, and prepare for fine and orderly electricity management.