Probabilistic prediction of ship destinations based on traffic pattern awareness in maritime networks

Abstract

Listen

Translate article

As maritime traffic becomes increasingly heavy, issues such as traffic congestion and inefficient transportation emerge as more prominent concerns. Ship destination prediction is essential for optimizing scheduling and improving maritime transportation efficiency. Currently, data quality and computational resources frequently limit the accuracy of most existing prediction methods. Therefore, this paper proposes a novel destination prediction method based on traffic pattern awareness. It uses AIS data to construct a maritime route network that integrates the navigational experience of ship groups. By modeling marine traffic patterns through abstract network topology and analyzing the target's real-time trajectory, the method employs a random forest-based prediction model to calculate the probabilities of candidate destinations and determine the final port. Experiments using 2020 Pearl River Estuary AIS data validated the proposed model. Through comparative analysis with four machine learning methods, the model demonstrated the highest accuracy and stable predictive performance. Additionally, the effectiveness of the model was further highlighted through multi-step dynamic predictions. At last, comparison with an advanced prediction model confirmed the proposed method's accuracy, demonstrating its ability to provide valuable decision-making support for ports in enhancing maritime traffic scheduling efficiency.