1.School of Electric Power Engineering, South China University of Technology, Guangzhou 510640, China;2.School of Automation Science and Engineering, South China University of Technology, Guangzhou 510640, China

At present, the research on spatial load forecasting lacks the consideration of complex spatial-temporal relationship. Therefore, a regional ultra-short-term spatio-temporal load forecasting model considering multi-dimensional and multi-source features is proposed in this paper. Firstly, based on the existing regional-level load, cell partitioning is carried out to construct a graph topology that considers cell correlation. Secondly, effective features are extracted from the spatial, feature, and temporal dimensions through the graph attention network, one dimensional convolutional network and gated recurrent unit, connecting the fully connected layers to output the results. Finally, simulation validation is conducted based on the real power load data from the New England region of the United States, and model attention weights are extracted to analyze the spatial dependencies between cells. The results show that, compared with the traditional models, the proposed model provides higher accuracy and stability with different prediction steps, effectively exploiting the spatial dependence of regional spatial load.