Abstract: In order to improve the prediction accuracy and robustness of non-stationary wind power series, a wind power prediction method based on principal component analysis (PCA) and series-core fused time series forecaster (SOFTS) is proposed in this paper. Firstly, PCA is used to calculate the principal components of wind power variables to construct model feature variables, which are input into the SOFTS multi-layer perceptron for wind power related variables sequence encoding. The core representation of each input sequence is obtained based on the spatiotemporal aggregation and redistribution module. Secondly, complex correlations and global feature information between channels are captured effectively through horizontal and vertical intersections. Finally, by training the SOFTS network parameters, a wind power prediction model based on PCA and SOFTS fusion is established. The actual operation data of a wind farm in a certain area is used for verification. The prediction accuracy of the proposed wind power prediction model is 98.38%. Compared with other models, it has higher prediction accuracy and stronger robustness to non-stationary sequences.