The overall power prediction of wind power clusters is of great significance for the optimal scheduling of regional wind power, especially at the present stage when extreme wind weather such as typhoons occurs frequently. The existing cluster prediction methods do not consider the differential fluctuation of the numerical weather forecast information of each wind farm in the cluster in the time series during the transit of typhoons, and the cluster is reasonably divided according to this. Therefore, this paper proposes a wind power cluster day-ahead power prediction method based on cluster dynamic partitioning and TCN-Transformer. First, the influence law of typhoon weather on the power generation of wind farm stations is revealed, and the key meteorological factors of typhoon weather affecting the wind power output are explored. On this basis, a cluster dynamic division(CDD) method of wind power clusters based on ISODATA algorithm is proposed during typhoon transit, and independent clustering classification is carried out for each period. In addition, TCN-Transformer neural network prediction method is introduced to conduct overall modeling for each subset group. Finally, the effectiveness of the proposed method is verified by a numerical example. The numerical results show that the proposed method has a certain degree of improved accuracy compared with the comparison method.