Wind Power Prediction Based on Spatiotemporal Informer and Transfer Learning

In order to solve the problem of low prediction accuracy caused by the instability of wind power series and the limited monitoring data of some wind farms, and fully consider the complex spatiotemporal characteristics of the coupling of different base stations in wind farms, a wind power prediction model based on graph convolution network and spatiotemporal feature interaction in Informer with transfer learning (GFIITL) is proposed. First, the correlation between the historical power of wind farms in the region is calculated, the feature adjacency matrix is constructed, and the node power information is updated through graph convolution network (GCN). Second, a Feature Interaction (FI) mechanism is proposed to extract information from multiple angles with dual‐channel input, and the model is migrated from the source monitoring station to the target station by using cyclic fine‐tuning transfer learning, so as to improve the prediction performance under the condition of limited historical data. Finally, compared with the traditional Informer model and other baseline prediction methods, the GFIITL model shows higher prediction accuracy and stability under the condition of limited data, which provides a new perspective for the research and application of wind power prediction.