End-to-end model-based trajectory prediction for ro-ro ship route using dual-attention mechanism

Area of Science:

• Maritime Logistics
• Artificial Intelligence
• Data Science

Background:

• Increasing global trade volume leads to shipping route congestion.
• Effective maritime service and management require accurate trajectory prediction.
• Existing prediction models face performance and generalization challenges.

Purpose of the Study:

• To propose a novel dual-attention (DA) based end-to-end (E2E) neural network (DAE2ENet) for enhanced trajectory prediction.
• To address the performance and generalization bottlenecks in current trajectory prediction models.
• To validate the effectiveness of DAE2ENet in a real-world maritime scenario.

Main Methods:

• Development of a dual-attention (DA) based end-to-end (E2E) neural network (DAE2ENet).
• Integration of Long Short-Term Memory (LSTM) units within the E2E architecture for sequential data processing.
• Implementation of global attention between encoder-decoder layers and multi-head self-attention for feature extraction.

Main Results:

• DAE2ENet demonstrated superior performance compared to baseline and benchmark models in trajectory prediction.
• The model effectively captured sequential features and interactions between input and output trajectories.
• Validation confirmed the model's ability to incorporate environmental factors influencing ship navigation.

Conclusions:

• The proposed DAE2ENet offers a significant advancement in trajectory prediction accuracy and model generalization.
• The dual-attention mechanism is crucial for improving the model's ability to handle complex sequential maritime data.
• DAE2ENet provides a robust solution for efficient maritime service and management amidst increasing shipping volumes.