Using Deep Learning to Forecast Maritime Vessel Flows.

Area of Science:

• Maritime intelligent transportation systems
• Deep learning applications in logistics
• Traffic flow prediction modeling

Background:

• Real-time maritime traffic data is crucial for efficient transportation systems.
• Accurate vessel flow information aids in congestion mitigation, emission reduction, and safety enhancement.
• Current methods may not fully capture the complexities of maritime traffic patterns.

Purpose of the Study:

• To propose and evaluate deep learning models for forecasting maritime vessel inflow and outflow.
• To compare the performance of Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and a hybrid Bidirectional LSTM-CNN (BDLSTM-CNN) model.
• To validate the effectiveness of these models using real-world Automatic Identification System (AIS) data.

Main Methods:

• Development of three deep learning architectures: CNN, LSTM, and BDLSTM-CNN.
• Spatial discretization of maritime regions into M x N grids for localized forecasting.
• Training and testing models on historical AIS data from Singaporean waters.

Main Results:

• All proposed deep learning models significantly outperformed conventional forecasting methods.
• The BDLSTM-CNN hybrid model demonstrated superior predictive accuracy.
• Performance was evaluated using standard metrics: Mean Absolute Error (MAE) and Root Mean Square Error (RMSE).

Conclusions:

• Deep learning approaches, particularly the BDLSTM-CNN, are highly effective for maritime vessel flow forecasting.
• The study validates the potential of these advanced models for enhancing maritime traffic management.
• Accurate forecasting can lead to improved operational efficiency and safety in maritime domains.