Abstract: With the increasing installed capacity of wind power, the impact of the power prediction error of wind farms and wind farm clusters on dispatching and operation is increasing. Current research on the characteristics of wind power prediction error is mainly for a single wind farm, which cannot meet the current demand of power grid dispatching and operation. Therefore, the paper counts and analyzes the distribution characteristics of a large number of wind farms, summarizes the variation law of distribution characteristics of the prediction error for different scales of wind farm clusters, and explores the best fitting model of the probability distribution of cluster prediction error. Then, in view of the guiding role of the prediction error distribution, the evaluation index for the persistence of error distribution is proposed. Finally, a more comprehensive analysis on the role of the prediction error research of wind power is carried out, and it is also pointed out that it has more engineering applicability to carry out the power grid dispatching with wind farm station clusters as a unit.