Spatial-Temporal Prediction of Schedulable Capacity of Electric Vehicles based on Graph Convolutional Network with Spatial-Attention

Spatial-Temporal prediction of the electric vehicles schedulable capacity (EVSC) is fundamental for their participation in various ancillary services in smart grid. This paper proposes a spatial-temporal prediction model based on graph convolution combining spatial-attention mechanism, focusing on the charging station level. Initially, each charging station within a region is considered a node, with historical charging data as node attributes. The adjacency matrix is determined based on the spatial distance between charging stations, forming the charging station connectivity graph. Subsequently, a combination of graph convolution and spatial-attention mechanism is employed to extract dynamic spatial features from the charging data. The temporal features of the charging data are then revealed using gate recurrent units, resulting in spatial-temporal predictions of the EVSC. The proposed method is validated using data from selected charging stations in a specific city and compared with other prediction methods. The prediction results show the proposed method can effectively explore the spatial-temporal features of charging station EVSC and improve the prediction accuracy.