Short-term power prediction method of wind farm cluster based on deep spatiotemporal correlation mining

This paper proposed a short-term power prediction method based on spatiotemporal correlation mining for wind farm clusters. Firstly, a quantitative metric for spatial correlation is established, which takes into account both wind speed and direction. Based on this metric, a graph structure that includes virtual nodes is constructed to represent the spatial correlation between wind farms, with the virtual nodes adding extra useful information to the input data. Then, we employ the graph attention network to extract the spatial features of the wind farm cluster, and then construct a bidirectional recurrent residual network to extract temporal features, introducing multi-task learning algorithms to optimize the network output. Lastly, an evaluation index for the false prediction component is proposed, which assesses the erroneous predictions caused by the accumulation of positive and negative errors, offering a reference for the development of power generation plans. Experimental analysis was conducted using data from 21 wind farm clusters in China, and the short-term prediction accuracy achieved was 89.69 %, which validated the effectiveness of the proposed model.