The invention relates to a user typical load curve acquisition method based on k-shape clustering, and belongs to the field of power system demand side response. The method comprises the steps of loadcurve normalization, load curve similarity definition, clustering center initialization, load curve classification, clustering center updating and classification convergence. The method is based on k-means clustering improvement, the cosine distance is used for replacing the Euclidean distance to serve as the standard for measuring the similarity between different load curves, the shape characteristics of the load curves can be better distinguished without being affected by amplitude fluctuation, and typical load characteristics of users can be extracted. Besides, compared with a traditionalclustering method using arithmetic mean of samples of the same type as a clustering center, the method obtains the clustering center by solving a maximum Rayleigh entropy model, so that the total similarity between the center and the samples of the same type is maximum, typical characteristics of a load curve can be better reflected, and accurate analysis of user loads is facilitated.