Aiming at the problems of poor stability of clustering results, poor effectiveness in clustering, slow speed and high memory consumption when making traditional clustering analysis for a large dimensionality huge number of load profiles with huge difference between the clusters under the background of the big data, an improved density peaks clustering algorithm is proposed. Firstly, principle components analysis method is used to reduce dimensions of load curves after normalization in order to reduce the calculation of the Euclidean distance between the sample vectors and to speed up the subsequent operations. Then, the kd tree algorithm is used to carry out the fast k-nearest neighbor search to generate KNN matrix. Finally, the KNN matrix is used to replace the original distance matrix as the input data. Based on the KNN improved local density and distance calculation criterion, the density peaks clustering algorithm is used to cluster the load profiles. Experiments and case analysis show that the proposed method is practicable and effective. This work is supported by National High-tech R & D Program of China (863 Program) (No. 2015AA050203).