Abstract: Aiming at the high-dimensional disaster of wind power prediction data and the communication dependence of cloud computing,a lightweight prediction method of wind power is proposed. KNN algorithm is used to calculate the distance quantization spatial correlation between fans,and the contour coefficient is introduced to adaptively determine the nearest neighbor number k to reduce the redundant feature dimension,so as to determine the adjacent fan data of input target prediction fan;GRU-MLP network based on Seq2Seq structure completes the wind power prediction of each fan. The experimental results show that the proposed method has less complexity and higher efficiency than the conventional network under the premise of approximate prediction accuracy,can provide technical solution for the migration of wind farm power prediction tasks from cloud to edge.