Short-term Wind Power Prediction Based on Improved Informer Incorporating Spatiotemporal Features

To address the low prediction accuracy caused by the instability of wind power series and the limitations of traditional prediction methods in capturing the spatiotemporal correlation between wind farms, a novel wind power prediction model was proposed. Firstly, by employing an improved embedding method, a temporal-feature hybrid multilayer perceptron (MLP) was proposed to enhance the ability to capture multiple correlations among feature dimensions and extract correlation information between global temporal and dynamic features. Then, the spatial feature interaction (SFI) mechanism was proposed. Utilizing dual-channel input capability, a feature extraction module based on attention mechanism was constructed to acquire dynamic spatial dependencies between the target wind farm and surrounding wind farms. Finally, the spatiotemporal feature map incorporating multiple wind farms was input into the encoder to obtain the prediction results. The proposed method was experimentally analyzed and verified by multiple sets of actual data. Results show that the proposed prediction method outperforms other prediction methods in terms of prediction accuracy.