Sign in Subscribe
Periodical publishing

Advancements in Marine Science and Engineering: Trajectory Prediction Based on Large Language Models

Strategic Insights

— 1 min read

Researchers from the National University of Defense Technology have published a new study on marine science and engineering, focusing on the deployment of trajectory prediction technology to ensure navigational safety in the rapidly expanding maritime trade industry. Inspired by the robust semantic comprehension exhibited by large language models (LLMs), the study introduces a novel trajectory prediction method leveraging LLMs to extract multidimensional semantic features of trajectories from comprehensive natural language narratives.

Key Takeaways:

  • The maritime transportation industry has experienced significant growth and complexity due to the expansion of maritime trade.
  • Current deep learning methods struggle to effectively integrate high-level semantic cues, such as vessel type, geographical identifiers, and navigational states, within predictive frameworks.
  • The proposed method leverages large language models (LLMs) to extract multidimensional semantic features of trajectories from comprehensive natural language narratives.
  • The LLM is fine-tuned using a supervised approach rooted in Low-Rank Adaptation (LoRA) to adapt to specific maritime areas or vessel classifications.
  • The model demonstrates notable performance in short-term predictions, with an average distance error of 5.26 nmi and 6.12 nmi for VTLLM and TrAISformer, respectively.
  • A performance improvement of approximately 14.05% is achieved compared to prevailing advanced models for ship trajectory prediction.
  • The study integrates ship identity, spatiotemporal trajectory, and navigational information through prompt engineering, enabling the LLM to extract rich semantic features from AIS data.

Statistics:

  • The maritime transportation industry has experienced burgeoning growth due to the expansion of maritime trade.
  • The proposed method leverages the robust semantic comprehension exhibited by large language models (LLMs) to extract multidimensional semantic features of trajectories.
  • The model demonstrates a performance improvement of approximately 14.05% compared to prevailing advanced models for ship trajectory prediction.
  • The averge distance error for VTLLM and TrAISformer are 5.26 nmi and 6.12 nmi, respectively.
  • The study was conducted by researchers from the National University of Defense Technology, including Ye Liu, Wei Xiong, Nanyu Chen, and Fei Yang.

Sources:

  • VTLLM: A Vessel Trajectory Prediction Approach Based on Large Language Models. Journal of Marine Science and Engineering, 2025, 13(9):1758. (Journal of Marine Science and Engineering - http://www.mdpi.com/journal/jmse)
  • MDPI AG. Journal of Marine Science and Engineering. 2025.
  • doi-org.sdpl.idm.oclc.org/10.3390/jmse13091758 (https://doi.org/10.3390/jmse13091758)

Read more