Abstract: The participation of power users in grid-side interactive power and auxiliary services has become a hot topic. Analysis of users interaction power behavior is a core task. Combining self-organizing map SOM neural network and K-means clustering algorithm， this paper uses a self-organizing center K-means algorithm for cluster analysis of users interaction electricity behavior and it can achieve more accurate recognition and fast clustering. Firstly， the principle of K-means algorithm in self-organizing center is analyzed， which shows its advantages in clustering analysis of electricity usage compared with traditional clustering algorithm. Then， under the background of peak-to-valley time-of-use electricity price， the adjustment potential index based on user psychology is constructed， and the cluster analysis of users electricity consumption behavior based on load data and adjustment potential index is analyzed. Finally， the daily load data of users in a jurisdiction of a power company is studied and the two-stage clustering results based on self-organizing center K-means algorithm are compared with the clustering results based on K-means algorithm， which proves the advantages of self-organizing center K-means algorithm based on adjustment potential index in the users accurate recognition and accurate clustering.

Key words:  user interaction, self-organizing center K-means algorithm, load data, adjustment potential indicator, clustering analysis