Abstract: Load clustering can help sort out load profiles according to their morphological difference, realize the analysis of users’ energy consumption behaviors, and provide important decision-making information for demand-side response, grid customer service and so on. An ensemble spectral clustering algorithm for load profiles considering dual-scale similarity is proposed. First, in order to overcome the limitation of Euclidean distance in measuring the similarity of load characteristics, the cosine distance of the load difference vector is used to measure the similarity of load shape changes, and a dual-scale similarity measure is proposed.Then, differential basic clustering models are established based on the dual-scale similarity and spectral clustering algorithm.Finally, the weights of basic clustering models are calculated adaptively according to clustering evaluation indices and ensemble clustering is constructed by means of weighted consistency matrix and spectral clustering. Numerical experiments prove that the proposed method can effectively mine the differences in load morphology characteristics and has more stable performance in different datasets, which verifies its good clustering effectiveness and robustness.