Due to the high randomness and irregularity of residential users’electricity demand, detailed data analysis is urgently needed to define the behavior characteristics of users to provide more reasonable electricity suggestions and demand response potential. Based on the fine-grained electricity consumption data and user information of residents, a classification of electricity residential users based on multi- dimensional electricity consumption behavior data is proposed. First of all, the non-intrusive smart meter is used to obtain the fine-grained electricity consumption data of residents；Then the user’s electricity consumption behavior is analyzed, and electricity consumption characteristics are found. Then,the CRITIC weight method is used to adaptively configure the weights of each index, and through the evaluation indicators of 6 types of clusters, 4 kinds of clustering algorithms and 3 data distance calculations are compared to achieve the optimal clustering method and the choice of the number of clusters. The actual data of a residential area are used to verify the power consumption characteristics and the effectiveness of the weight-fixing clustering method proposed in this paper, and the residential user groups are divided into two categories with obvious differences in electricity consumption behavior.