A Ship Trajectory Prediction Framework Based on a Recurrent Neural Network.

Area of Science:

• Maritime Navigation and Safety
• Artificial Intelligence in Transportation
• Geospatial Data Analysis

Background:

• Accurate ship trajectory prediction is crucial for maritime safety and early warning systems.
• Existing methods face challenges in balancing prediction accuracy with computational efficiency.
• Automatic Identification System (AIS) data provides valuable information for vessel tracking.

Purpose of the Study:

• To develop and evaluate a deep learning framework for enhanced ship trajectory prediction.
• To improve the computational efficiency of vessel trajectory prediction models.
• To validate the proposed model using real-world maritime data.

Main Methods:

• Extraction of ship trajectories from AIS data (longitude, latitude, speed, course).
• Application of Density-Based Spatial Clustering of Applications with Noise (DBSCAN) for trajectory derivation.
• Implementation of a trajectory information correction algorithm using symmetric segmented-path distance.
• Development and training of a recurrent neural network, specifically a Gate Recurrent Unit (GRU) model.
• Validation using ground truth AIS data from Zhangzhou port, China.

Main Results:

• The proposed GRU model demonstrates improved computational time efficiency for ship trajectory prediction.
• Prediction accuracy of the GRU model is comparable to that of the Long Short-Term Memory (LSTM) network.
• The trajectory information correction algorithm effectively optimizes incoming trajectories by reducing redundant data.

Conclusions:

• The developed deep learning framework, utilizing a GRU model, offers a computationally efficient solution for real-time ship trajectory prediction.
• The method effectively processes and optimizes AIS data for improved prediction outcomes.
• This approach contributes to enhancing maritime navigation safety and early warning capabilities.