[Paper Review] The Unsupervised Method of Vessel Movement Trajectory Prediction

Chih‐Wei Chen, Charles Harrison+1 more|arXiv (Cornell University)|Jul 27, 2020

Maritime Navigation and Safety9 references4 citations

TL;DR

This paper proposes an unsupervised clustering-based trajectory reconstruction (CBTR) method for predicting vessel movement trajectories using AIS data without requiring training data. By modeling vessel motion in a 3D feature space of time difference, rescaled error distance, and space-time angle, the method improves prediction accuracy by selecting path-fitting candidates, achieving over 99% correct-neighbor rates even under 50% data down-sampling.

ABSTRACT

In real-world application scenarios, it is crucial for marine navigators and security analysts to predict vessel movement trajectories at sea based on the Automated Identification System (AIS) data in a given time span. This article presents an unsupervised method of ship movement trajectory prediction which represents the data in a three-dimensional space which consists of time difference between points, the scaled error distance between the tested and its predicted forward and backward locations, and the space-time angle. The representation feature space reduces the search scope for the next point to a collection of candidates which fit the local path prediction well, and therefore improve the accuracy. Unlike most statistical learning or deep learning methods, the proposed clustering-based trajectory reconstruction method does not require computationally expensive model training. This makes real-time reliable and accurate prediction feasible without using a training set. Our results show that the most prediction trajectories accurately consist of the true vessel paths.

Motivation & Objective

• Address the challenge of predicting vessel trajectories in real-time using sparse, noisy, and irregularly sampled AIS data without labeled training data.
• Overcome limitations of supervised learning models in handling variable sample sizes, irregular time-sampling, and high noise in maritime AIS streams.
• Develop a scalable, big-data-compliant method that reconstructs vessel tracks by leveraging local positional and kinematic information.
• Enable accurate trajectory prediction under data gaps and missing identities, relevant for maritime security and navigation.
• Demonstrate the method’s robustness on real and simulated regional AIS data from Norfolk, Virginia, under down-sampling conditions.

Proposed method

• Represent each vessel point in a 3D feature space combining time difference between points, rescaled error distance between predicted and actual positions, and space-time angle.
• Use forward and backward predicted positions (P⁺(xᵢ) and P⁻(xⱼ)) to compute error distance, measuring path fit rather than static point proximity.
• Apply clustering via local optimization: connect points xᵢ and xⱼ if the error distance is minimized and angular constraints are satisfied.
• Enforce geometric and temporal constraints: time difference ≤ 1000 seconds, space-time angle ≤ 84.26° (for moving pairs) or ≤ 18.19° (for steady pairs), and spatial distance < 350 meters.
• Use threshold-based selection for rescaled error distance and turning angle (≤ 53.13°) to filter plausible trajectory segments.
• Iteratively merge candidate points into coherent tracks using a greedy, local search strategy that avoids global model training.

Experimental results

Research questions

• RQ1 Can an unsupervised method reconstruct vessel trajectories accurately from noisy, incomplete AIS data without any labeled training data?
• RQ2 How effective is the 3D feature space (time difference, rescaled error distance, space-time angle) in improving trajectory prediction accuracy compared to conventional methods?
• RQ3 To what extent does the proposed CBTR algorithm maintain accuracy under data down-sampling (e.g., 20% and 50% point removal)?
• RQ4 What are the sufficient and necessary conditions for correct clustering of vessel paths using the proposed method?
• RQ5 How does the method perform in real-world regional maritime data with irregular sampling and missing identities?

Key findings

• The proposed CBTR method achieves correct-neighbor rates of 0.9960, 0.9951, and 0.9942 for AIS datasets down-sampled by 20% using method (1), indicating high robustness.
• Under 50% down-sampling (method 1), correct-neighbor rates remain high at 0.9924, 0.9910, and 0.9892, demonstrating resilience to data gaps.
• The method successfully reconstructs true vessel paths without training, making it suitable for real-time deployment in operational maritime environments.
• Theoretical analysis confirms that correct path clustering occurs if and only if specific geometric and temporal conditions (e.g., angle ≤ 84.26°, time gap ≤ 1000 s) are met.
• The method outperforms traditional approaches by focusing on predicted path fit rather than static point proximity, reducing false connections.
• The algorithm maintains high accuracy even when data are sparse or corrupted, due to its reliance on local kinematic consistency and predictive modeling.