GHP-ManageFormer: A Transformer-Enhanced Approach for Logistics Transportation Time Prediction

This paper proposes a Transformer-based deep learning model, GHP-ManageFormer, for solving the problem of logistics and transportation timeliness prediction. The model adopts a multi-layer architecture, which processes static, temporal, and dynamic features in logistics and transportation through feature processing and attention layers. The core innovation of the model lies in the introduction of a spatiotemporal-aware multi-head attention mechanism and a cross-modal integration strategy to achieve effective fusion of multi-source heterogeneous data. Experiments based on a logistics dataset containing more than 100,000 delivery route records show that GHP-ManageFormer outperforms existing methods in terms of both the root mean square error (RMSE=1.240) and the coefficient of determination (R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>=0.856), significantly improving the performance of the Transformer benchmark model. In addition, the model has the ability to estimate the uncertainty of the prediction results, providing more comprehensive reference information for practical logistics decisions.