Abstract: Accurate wind speed prediction is essential for the stable and economical operation of the power system, and an ultra-short-term prediction model of convolutional memory network is proposed based on information aggregation of cluster space decoupling. Firstly, the influence of the wake effect of the cluster is analyzed, and the wake effect impact factor is embedded into cluster analysis to realize the space decoupling based on the wake correlation of the wind turbines. Then, the spatio-temporal correlation index is constructed. The representative wind turbine is selected from each decoupling cluster. And the spatio-temporal information domain is extended by combining temporal information similarity. Based on the aggregation information of the high-order spatiotemporal domain, the convolutional memory network is constructed to enhance the spatiotemporal characteristics and carry out ultra-short-term wind speed prediction. Finally, the proposed model is applied to the wind speed prediction of actual wind farm, the effectiveness and applicability of the model are verified through comparative analysis.