Abstract: | Addressing the limitation of traditional charging station load prediction methods, which only forecast the load prediction of a single site, the paper proposes a collaborative forecasting method for multiple charging stations using the Graph Spatiotemporal Neural Network (GSTNN). Firstly, a spatiotemporal infographic is defined to describe the spatiotemporal relationship between charging station loads. Then, a spatiotemporal feature extraction network is constructed. It utilizes the graph convolutional neural network and the gated sequence convolutional network to extract spatial and temporal dimension information from the infographic. Furthermore, the Long Short-Term Memory Networks (LSTM) are used to mine external feature information that affects load prediction. Finally, all the extracted features are fused to predict the load.The results from the test cases show that the method based on the GSTNN model fully considers the influences of spatiotemporal characteristics and external features, cooperates with the load data of multiple charging stations for prediction, and produces results for each station concurrently, thereby effectively impoving prediction accuracy and supporting the stable operation of the power grid. |