Online Prediction of Ship Behavior with Automatic Identification System Sensor Data Using Bidirectional Long...

Area of Science:

• Maritime technology
• Artificial intelligence
• Data science

Background:

• Real-time ship behavior prediction is crucial for navigation safety and collision avoidance.
• Existing models may not fully capture complex sequential data from Automatic Identification System (AIS).

Purpose of the Study:

• To develop an online, real-time ship behavior prediction model using a BI-LSTM-RNN.
• To leverage AIS data's sequential characteristics for improved prediction accuracy and online parameter adjustment.

Main Methods:

• Constructed a bidirectional long short-term memory recurrent neural network (BI-LSTM-RNN).
• Utilized the "forget gate" mechanism within LSTM units to manage historical and unique behavioral patterns.
• Trained the model using 2015 AIS data from Tianjin Port.

Main Results:

• The BI-LSTM-RNN model demonstrated effective prediction of ship navigational behaviors.
• The bidirectional structure improved data relevance, enhancing prediction accuracy.
• The "forget gate" improved model universality by retaining common patterns and discarding unique ones.

Conclusions:

• The developed BI-LSTM-RNN model offers a robust solution for real-time ship behavior prediction.
• This advancement contributes to increased efficiency and safety in maritime operations.
• The model serves as a predictive foundation for intelligent collision avoidance, route planning, and anomaly detection systems.