Anomaly detection of ship behavior based on deep neural networks

•A HybridAttn-BiRNN model with temporal and feature attention is proposed for ship trajectory prediction.•An anomaly detection framework integrating HDBSCAN clustering and the HybridAttn-BiRNN model is established.•Multiple anomaly types are detected using threshold discrimination and feature deviat...

Full description

Saved in:
Bibliographic Details
Published in:Reliability engineering & system safety Vol. 266; p. 111801
Main Authors: Shu, Yaqing, Dong, Ao, Liu, Chengyong, Gan, Langxiong, Song, Lan
Format: Journal Article
Language:English
Published: Elsevier Ltd 01.02.2026
Subjects:
AIS data
Anomaly detection
Deep neural networks
HDBSCAN
Ship behavior
HDBSCAN
AIS data
Deep neural networks
Anomaly detection
Ship behavior
ISSN:0951-8320
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:•A HybridAttn-BiRNN model with temporal and feature attention is proposed for ship trajectory prediction.•An anomaly detection framework integrating HDBSCAN clustering and the HybridAttn-BiRNN model is established.•Multiple anomaly types are detected using threshold discrimination and feature deviation analysis.•Ship anomalies are detected with 92.7 % accuracy using Shanghai Port AIS data in experiments. With the acceleration of the digitalization process in the global shipping industry. The anomaly detection of ship behaviors, including ship speed, course, and trajectory, has become a core challenge to ensure maritime traffic safety. To address this challenge, a method that integrates a hierarchical density-based spatial clustering algorithm (HDBSCAN) with a deep learning model is proposed in this paper. Unlike conventional approaches that rely solely on clustering or prediction models, AIS trajectories are compressed and clustered by our method to extract normal behavior patterns. Ship behavior is then predicted using a deep learning model with feature and temporal attention mechanisms. A dual-threshold mechanism is employed to identify various anomalies, including speed and course deviations, based on prediction errors. A case study is conducted using the AIS data from the waters of Shanghai Port. The experimental results show that abnormal ship behavior can be effectively identified by this method, with an average detection accuracy of 92.7% for various types of anomalies. The method proposed in this research could proactively identify potential maritime risks, thereby improving the overall safety and reliability of shipping operations.
ISSN:0951-8320
DOI:10.1016/j.ress.2025.111801