Through the classification and analysis of load data, the behavior characteristics of power users can be obtained,which provides support for the formulation of demand response strategies and effect evaluation. Firstly, the load data is preprocessed including identifying and processing abnormal numbers,and smoothing to remove burr data. Secondly, with the reason that the fuzzy C-means clustering algorithm is sensitive to the initial cluster center, easy to fall into local optimality, and is greatly affect-ed by noise, the shortest distance clustering method is proposed to provide the initial clustering center for fuzzy C clustering. The efficiency index is used to select the best clustering results of different categories, and the data density is used to identify and eliminate noise points.Finally, through comparison with other methods and cluster analysis of a certain hemp spinning enterprise’s load, the correctness and effectiveness of the improved algorithm is verified.