Predicting inland waterway freight demand with a dynamic spatio-temporal graph attention-based multi attention network

Inland waterway transport (IWT) has gained significant attention for its environmental sustainability. Consequently, there is an increasing focus on boosting IWT’s market share to reduce transportation emissions. Accurate forecasting of IWT freight demand is crucial for ports to plan long-term targets and support a mode shift towards sustainable transport. However, forecasting IWT demand is challenging due to the complexity of external environments. This paper introduces a Dynamic Graph Attention Multi-attention Network (DGAT-MAN) model designed to forecast IWT freight demand by capturing evolving spatial and temporal dynamics. Our comparative analysis demonstrates that this model significantly outperforms established baseline approaches. As one of the first studies to apply spatio-temporal deep learning models to IWT demand forecasting, this work contributes a novel approach to enhancing sustainable transport planning. • Propose a customized graph deep learning model for accurate inland waterway transport demand. • Combine temporal data with both dynamic and static spatial features for water transport demand forecasting. • Provide decision-support tools for strategic planning and sustainability in logistics. • Experimental results demonstrate that our proposed model outperforms benchmark methods.