Cluster-aware deep learning-based vessel trajectory prediction for maritime traffic management

Li, Yan, Song, Lan, Yang, Zaili and Li, Huanhuan (2026) Cluster-aware deep learning-based vessel trajectory prediction for maritime traffic management. Ocean & Coastal Management, 276, [108160]. (doi:10.1016/j.ocecoaman.2026.108160).

Abstract

AbstractAccurate vessel trajectory prediction is essential for maritime safety and traffic management. While Automatic Identification System (AIS) data offer valuable spatiotemporal information, reliable forecasting remains challenging due to heterogeneous traffic patterns and dynamic behaviors. This study proposes a two-stage, cluster-aware framework, IDCMSA-MSIFE, that integrates trajectory clustering with multi-source deep learning to enhance prediction accuracy and robustness. In the first stage, an Improved Density Clustering with Mapping and Spatial Accessibility (IDCMSA) method is developed to identify representative traffic patterns by mapping trajectories to characteristic points and detecting their density-reachable relationships within a grid-based structure. This process distinguishes heterogeneous navigation behaviors in complex maritime environments. In the second stage, trajectories in each cluster are fed into a Multi-Source Information Fusion Enhanced (MSIFE) deep learning network. The model integrates multiple AIS-derived features, including trajectory coordinates, Course Over Ground (COG), and Speed Over Ground (SOG), through a self-attention mechanism, while Bidirectional Long Short-Term Memory (Bi-LSTM) units capture temporal movement dynamics. To enhance training stability and predictive consistency, correction terms are added to the loss function. Experimental results on real-world AIS datasets from two maritime regions show that our framework outperforms seven benchmarks in accuracy and robustness, demonstrating its value for navigation support and collision risk mitigation. This study provides a data-driven approach for safer maritime management.

This record has no associated files available for download.

Contributors

Author: Huanhuan Li

Download statistics