Abstract: The load model of electric power system is a key factor to determine the reliability of power system, and conventional clustering algorithm of load characteristic data suffers such disadvantages as high computing complexity and long operation time. Therefore, a novel distributed clustering model is built by organically combining K-means and Canopy clustering algorithms. Based on this, users integral point load data are applied with normalization processing, the clustering center distribution can be obtained by using the relationship between load normalized interval values and parameters at 24 integral time points. Taking historical day load data of Fujian province as example, the operation rapidity of distribution cluster model is verified. Results show that distance threshold T₂ is inversely proportional to the operation time, the more the number of clusters is the longer the operation time will be. In addition, the clustering center distribution of large-scale industry is stable, its significance is not obvious, while the significance of clustering center distribution of agricultural industry is obvious, which provide ideal references for predicting users load characteristics and power consumption features.