Spatiotemporal Forecasting for Regional Express Pickup Demand

To improve the efficiency of the parcel pickup operations, accurately identifying high-demand areas has become a key challenge. Based on point-of-interest data, the hierarchical graph infomax (HGI) method is employed to generate regional embeddings. These embeddings, along with historical pickup demand series, are then input into a graph multi-head attention network (GMAN) for prediction, thereby forming the proposed HGI-GMAN model. Experimental results on a real-world dataset from an express company indicate that the HGI-GMAN model outperforms five classical baseline models across various regression (RMSE, R²) and classification (macro-F1) metrics. Additionally, hyperparameter sensitivity analysis and ablation studies verify the robustness of the proposed model and the effectiveness of the extracted features.