Knowledge transfer strategy for enhancement of ship maneuvering model

Abstract

The advancement of autonomous vehicles and concerns about ship navigation safety have resulted in a greater need for ship model quality. However, in situations where there is limited prior information or data available about the system, the development of accurate models can be challenging. To address this issue, we propose a knowledge transfer strategy that can migrate and adapt domain knowledge from a well-modeled benchmark ship to a target ship. The benchmark, or source ship, should resemble the target ship in the feature space and reveal similar trends in the prediction horizon. By incorporating informative trends into the data-driven transfer function, the representative model of the target ship can be considerably enhanced. In this study, the experiments are conducted on two full-scale vessels that characterize different dimensions and dynamic properties. A feature vector is introduced to evaluate the configuration similarity between vessels, and ship maneuverability is compared to authorize the security of predictive tendency. The derived target ship model is verified to accurately predict maneuvering trajectories in various scenarios, demonstrating that knowledge transfer from a source ship facilitates the target ship modeling process. This approach provide new insights into the development of models for systems with confined information.