Predictive Clustering of Vessel Behavior Based on Hierarchical Trajectory Representation

TL;DR

Experiments conducted on real AIS datasets demonstrate that PC-HiV effectively captures behavioral evolution discrepancies between different vessel types and near emission control area boundaries, and proves the superiority of the proposed PC-HiV over existing models.

Abstract

Vessel trajectory clustering, which aims to find similar trajectory patterns, has been widely leveraged in overwater applications. Most traditional methods use predefined rules and thresholds to identify discrete vessel behaviors. They aim for high-quality clustering and conduct clustering on entire sequences, whether the original trajectory or its sub-trajectories, failing to represent their evolution. To resolve this problem, we propose a Predictive Clustering of Hierarchical Vessel Behavior (PC-HiV). PC-HiV first uses hierarchical representations to transform every trajectory into a behavioral sequence. Then, it predicts evolution at each timestamp of the sequence based on the representations. By applying predictive clustering and latent encoding, PC-HiV improves clustering and predictions simultaneously. Experiments on real AIS datasets demonstrate PC-HiV's superiority over existing methods, showcasing its effectiveness in capturing behavioral evolution discrepancies between vessel types (tramp vs. liner) and within emission control areas. Results show that our method outperforms NN-Kmeans and Robust DAA by 3.9% and 6.4% of the purity score.

Figures (8)

• Figure 1: Behavior sequence in different levels. (a) shows the mooring behavior sequence of different vessels over the same time period. The green trajectory of the passenger vessel indicates that it has made multiple trips between specific ports. The yellow trajectory of the cargo vessel indicates that it only made one call at port C during this period; The red trajectory of the tanker indicates that it will moor in specific ports (E, D) with the berth of the tanker. The behavior evolution of the same type of vessels is similar based on the frequency of mooring, port preference, etc.. (b),(c) divide vessel trajectory into different stages according to the timestamp $t$ of the behavioral transition point. (b) The vessel turns left, slows down to avoid hitting the reef, and then goes straight. (c) Two vessels meet, according to the COLREGSref5 rules, they will generally turn right to give way. As such, vessels have similar behavior sequences during the sailing process.
• Figure 2: Hierarchical vessel trajectory structure. One can observe a mapping relationship from the continuous position sequence to the sub-trajectory sequence, as well as a mapping relationship from the sub-trajectory sequence to the label sequence. Each segment in trajectory represents a behavior. In addition, the specific behaviors in the sub-trajectory are concerned and the sub-trajectory is mapped to a label point to form a label sequence.
• Figure 3: PC-HiV model framework. It consists of a vessel trajectory hierarchy representation module, label generation module, encoder, predictor, and assigner.
• Figure 4: Evaluation of vessel sub-trajectory sequence clustering results with different K
• Figure 5: Phased evolution of vessels in the same cluster