A graph attention network-gated recurrent unit based method for electric vehicle charging load prediction with multi-source heterogeneous feature integration
WEN Changbao, WU Benhuang, SUN Jieru
School of Energy and Electrical Engineering, Chang’an University, Xi’an 710064
Abstract：To mitigate the decline in prediction accuracy caused by the limited diversity of input features and the insufficient extraction of spatiotemporal correlations in existing electric vehicle (EV) charging load forecasting models, a novel spatiotemporal forecasting framework is proposed, in which a graph attention network (GAT) is integrated with a gated recurrent unit (GRU) to effectively capture complex spatial and temporal dependencies. The model constructs a graph structure based on geographical proximity, incorporating diverse features such as historical load, weather conditions, calendar dates, and holidays. A multi-head GAT is employed to extract spatial dependencies, while the GRU models temporal dynamics. The final prediction is generated through a fully connected layer. Experimental results demonstrate that the proposed method significantly outperforms traditional methods and mainstream deep learning approaches in terms of forecasting accuracy, while also maintaining robust performance in cross-regional scenarios. This method offers data support for the dynamic scheduling of urban power grids and the orderly management of EV charging.
2026, Vol. 27 
