Session: 06-05-04 Marine Hydrodynamics - IV

Submission Number: 181755

A Physics-Informed Data-Driven Model for Ship Maneuvering Motion Prediction 

Testing autonomous navigation algorithms for the unmanned surface vessel (USV) is time-consuming and costly. Virtual testing through simulation platforms can significantly reduce these expenses. Accurate modeling of ship maneuvering motion is essential for the simulation testing of autonomous navigation algorithms. This paper proposes a novel physics-informed data-driven model for ship maneuvering motion prediction. Instead of using conventional maneuvering test data, such as turning test and zigzag test, this method utilizes routine operational data from the vessel as the training set, thereby providing the model with more diversely distributed training samples. To enhance the model's generalization and robustness, virtual datasets conforming to the principles of ship kinematics and dynamics are incorporated to impose stronger constraints on the model. Although these datasets represent motion states typically unattainable for the vessel, they facilitate the model's deeper understanding of the ship's motion mechanisms. Two types of simulation tests were conducted based on the trained model. The first was a maneuvering test, where the introduction of random noise demonstrated the model's strong robustness. The second was a synchronous simulation using control signals from a real dataset, and the comparison with actual motion states validated the model's accuracy. This approach offers new insights into the data-driven modeling of ship maneuvering motion.

Presenting Author: Lei Dong Shanghai Jiao Tong University

Presenting Author Biography: Lei Dong is a Ph.D. candidate at the School of Naval Architecture, Ocean and Civil Engineering, Shanghai Jiao Tong University, China. His research primarily focuses on autonomous navigation algorithms for unmanned surface vessels (USVs) and data-driven modeling of ship maneuvering motion. He received his Bachelor of Engineering degree from Huazhong University of Science and Technology. His recent work includes the development of novel attention-based neural networks for ship motion prediction, as presented in his publications such as "An attention mechanism model based on positional encoding for the prediction of ship maneuvering motion in real sea state."

Authors:

Lei Dong Shanghai Jiao Tong University
Hongdong Wang Shanghai Jiao Tong University