The invention belongs to the field of simulation of electric power systems, and particularly relates to a cluster classification for a wind power plant. Clusters are classified in a unit of the wind power plant according to the actually measured operating data of the wind power plant. In the process of acquiring the data, the actually measured data probably contain noise data because of the factors like the defect or the execution error of a measurement system. In order to reduce the interference of the noise data, the isolated point data in the actually measured operating data of the wind power plant are firstly processed according to the potential value of a sample point. When the central initial positions of the two clusters are nearer during the cluster classification, more redundant information is contained, and the classification result easily becomes the locally best. Aiming at the problem, a sample group with the smallest Euclidean distance moves towards the mean value point, the mean value of the moved sample group replaces the original sample group, so that the method acquires the central position of the diversified initial clusters, and the global searching ability is improved. By the adoption of the cluster classification for the wind power plant, provided by the invention, wind turbine generators having the near operating points are classified in the same cluster, and the equivalent modeling approach for the wind power plant is optimized.