Electric Vehicle Charging Load Prediction Model Based on Improved PageRank and Spatiotemporal Convolutional Network

The existing prediction methods do not take into account the spatial dimension information of charging stations, resulting in limited prediction accuracy. This research introduces a novel approach for forecasting charging load patterns that integrates spatial dimension information. The improved PageRank algorithm evaluates node significance by incorporating historical charging load data and geographic positioning details of charging stations, enabling the extraction of initial spatiotemporal characteristics The multi-head attention mechanism is combined with the spatiotemporal graph convolutional network to further explore the complex associations and spatiotemporal features between nodes, and the dilated causal convolution structure significantly improves long-term dependency modeling capabilities, and fully realizes charging load prediction; experiments show that compared with the existing methods, the model reduces the MAE and RMSE indicators by $\mathbf{1 0. 5 6 \%}$ and 3.74% respectively, and demonstrates superior performance in forecasting scenarios characterized by significant load variations and extended temporal sequences, confirming that incorporating spatial dimension data effectively enhances charging load prediction accuracy.