[Paper Review] Modeling Historical AIS Data For Vessel Path Prediction: A Comprehensive Treatment

Enmei Tu, Guanghao Zhang+3 more|arXiv (Cornell University)|Jan 2, 2020

Maritime Navigation and Safety36 references31 citations

TL;DR

The paper proposes a comprehensive framework to model massive historical AIS trajectory data for vessel path prediction, combining outlier removal, concise feature representation, and motion-trend ensemble learning, and demonstrates superior accuracy versus several baselines.

ABSTRACT

The prosperity of artificial intelligence has aroused intensive interests in intelligent/autonomous navigation, in which path prediction is a key functionality for decision supports, e.g. route planning, collision warning, and traffic regulation. For maritime intelligence, Automatic Identification System (AIS) plays an important role because it recently has been made compulsory for large international commercial vessels and is able to provide nearly real-time information of the vessel. Therefore AIS data based vessel path prediction is a promising way in future maritime intelligence. However, real-world AIS data collected online are just highly irregular trajectory segments (AIS message sequences) from different types of vessels and geographical regions, with possibly very low data quality. So even there are some works studying how to build a path prediction model using historical AIS data, but still, it is a very challenging problem. In this paper, we propose a comprehensive framework to model massive historical AIS trajectory segments for accurate vessel path prediction. Experimental comparisons with existing popular methods are made to validate the proposed approach and results show that our approach could outperform the baseline methods by a wide margin.

Motivation & Objective

• Motivate and address the challenge of predicting vessel paths from irregular, high-volume AIS data with low quality.
• Develop a three-part framework: automatic trajectory outlier detection, concise sample representation, and motion-trend ensemble learning.
• Demonstrate robustness and accuracy on a real AIS dataset spanning multiple vessel types and regions.

Proposed method

• Detect trajectory outliers by identifying anomalous motion patterns (sharp turning and self-crossing) and applying an improved Ramer-Douglas-Peucker-based approach.
• Construct uniform, informative feature vectors from irregular AIS samples by transforming to a local coordinate system and extracting kinematic/static features (lat/long, velocity, SOG, COG, etc.).
• Implement motion trend ensemble learning by clustering samples, training cluster-specific models (ELM) for each cluster, and fusing predictions from the best-performing models via a weighted scheme.

Experimental results

Research questions

• RQ1 Can historical AIS trajectory segments be effectively modeled to predict future vessel positions over 15–60 minutes?
• RQ2 Does an ensemble of motion-pattern-specific models outperform single-model baselines on diverse AIS data?
• RQ3 How do sample representation and outlier removal influence predictive accuracy and robustness across vessel types and regions.

Key findings

• The proposed method achieves substantially lower prediction errors than baselines (ELM, LS-SVM, MLP, GMM, Gaussian Process Regression) across 15–60 minute horizons.
• The method yields lower standard deviation in prediction errors, indicating more stable performance.
• Including static information (vessel type, region) further improves prediction accuracy.
• An AIS database of around 100 GB with 200 trajectories from 180 vessels was used for evaluation.
• The approach outperforms baselines by a wide margin on the real-world AIS dataset.