Bridging similar ships’ dynamics for safeguarding the system identification of maneuvering models
Motoyasu Kanazawa, Tongtong Wang, Yasuo Ichinose, Robert Skulstad, Guoyuan Li, Houxiang Zhang
Abstract
System identification (SI) approach identifies ship maneuvering models using data from free-running maneuvers. As it is impossible to comprehensively validate model performance, a reliable model must be built within a consistent range with similar ships’ models. However, in the SI approach, models are greatly influenced by the dataset design and feature selection, leading to the instability of model identification and the failure to ensure such consistency. To address this issue, our new idea introduces similar ship’s model into model identification as knowledge foundation. First, we select a similar ship with three-step procedure, then build a new model by refining model parameters for a similar ship. Such a refinement is conducted with additional l2 regularization term in ridge regression, which balances knowledge and data-driven refinement. Designing the safe range of such refinement with hyperparameters, this study helps designers find a robust-and-accurate model within designed safe zones. In simulation experiments, we built knowledge connection between 161 m and 175 m container ships. By using the well-validated model of the former, a robust-and-accurate model for the latter was easily built by using a limited dataset, resulting in excellent model performance. This study makes the SI approach more promising by incorporating knowledge connections between similar ships.