A deep learning method to predict ship short-term trajectory for proactive maritime traffic management
Quandang Ma, Zhouyu Lian, Xu Du, Yuting Jiang, Ahmad BahooToroody, Mingyang Zhang
Abstract
With the rapid growth of global maritime trade and the increasing density of vessel traffic, the risk of ship collisions has become a growing concern, especially in busy and complex waterways. To support proactive maritime traffic management and enhance navigational safety, this paper presents a deep learning-based approach for short-term ship trajectory prediction. Specifically, we propose a multi-scale fusion Temporal Convolutional Network (TCN) that learns vessel movement patterns using data from the Automatic Identification System (AIS). The model captures key motion features—such as trajectory changes, speed, acceleration, turning angles, and rotation rate—to better reflect the dynamic behavior of ships. By combining short-term variations with longer-term trends, the proposed TCN model achieves more accurate and reliable predictions. Experiments on real AIS datasets demonstrate that our method outperforms existing techniques in both accuracy and robustness, particularly in complex coastal environments. This research contributes to smarter traffic control and safer maritime navigation, and supports the development of intelligent maritime systems.