Ultra-short-term Wind Power Prediction Based on Spatiotemporal Attention Convolution Model

With the continuous improvement of wind power utilization, accurate prediction of the wind power output power is of great significance for the scheduling and stable operating of the power systems. However, the randomness and volatility of the wind power generation easily affects the accuracy of the power prediction results. In this paper a wind power prediction based on the spatiotemporal correlation is proposed, consisting of a spatiotemporal attention module and a spatiotemporal convolution module. First, the spatial attention layer and the temporal attention layer are used to aggregate and extract the spatiotemporal correlations between different wind turbines. Second, the spatial features and the temporal evolution patterns among the wind power data are effectively captured by the spatial convolution layer and the temporal convolution layer. Finally, the prediction method is experimentally validated using the operational data from two actual wind farms in China. The results indicate that compared to the traditional prediction methods, the fusion of the spatiotemporal attention and the spatiotemporal convolution enables the proposed prediction to have a higher accuracy and a better stability. © 2024 Power System Technology Press. All rights reserved.