In order to accurately obtain representative wind power output scenarios, this paper proposes a typical wind power scenario clustering method based on the improved density peak optimization global K-means algorithm. Firstly, the density peak clustering algorithm (DPC) is used to preliminarily determine the cluster center, and then the global K-means (GKM) is used for iterative clustering. Subsequently, a zebra optimization(ZO) method with global search capability is introduced to find the optimal value of the truncation distance in the DPC. Then, in order to comprehensively evaluate the stability and effectiveness of the obtained scenario set, three clustering effectiveness indexes are selected and the entropy weight Topsis method is used for comprehensive evaluation. Finally, a large number of simulation experiments and analyses are carried out using actual wind power data. The results show that the proposed method has better DBI, SSE and SC values, and can more accurately extract and divide typical wind power scenarios.