A ship trajectory prediction model integrating ship-shore speech communication for early prediction at waterway intersections

Early ship trajectory prediction improves traffic coordination but increases the risk of intent misjudgment at waterway intersections, leading to deviations between predicted and actual trajectories. To address this, we propose a ship trajectory prediction model grounded in the International Maritime Organization (IMO) framework and the rule of “intent report - ship maneuver - trajectory change” observed in real-world waterway intersections. Our method enables early intent recognition by leveraging intent information embedded in ship-shore speech communication. Within a defined spatiotemporal range, we associate communication data with observed trajectories to identify reported intentions. The extracted intent labels are integrated with encoded historical trajectory features and fed into a decoder, dynamically constraining predicted directions. This alignment with reported intent advances the prediction timeline without compromising accuracy. Empirical validation at the Wusongkou Estuary (Shanghai, China) demonstrates that our model advances the prediction timeline by 6.94–8.4 min compared to existing models, while maintaining similar accuracy. This work pioneers the integration of ship-shore speech communication into trajectory prediction, highlighting the potential of AI-driven maritime safety systems.