Short-term Charging Quantity Prediction of Charging Stations Based on Graph Neural Networks

With the rapid expansion of the Chinese electric vehicle (EV) industry, the ever-increasing number of EVs and their supporting charging infrastructure pose significant challenges to the national power grid. To tackle these challenges, accurately predicting short-term charging demand at EV stations is vital for strategic infrastructure construction, grid optimization, and sustainable EV development. This study assesses various sequential data prediction models, primarily exploring graph neural network (GNN) approaches, using authentic historical charging data from urban EV stations. By incorporating multi-graph and dynamic factors, DM-STGCN(A) effectively tackles the complexities of spatiotemporal dependencies in charging demand prediction, demonstrating exceptional accuracy with MAE of 0.195, MAPE of 18.019%, and RMSE of 0.217 across multiple performance metrics. Based on this model, a robust short-term charging demand prediction system is designed to empower operational decisions and optimize charging infrastructure.