Ship manoeuvring model identification using mutual information clustering with stepwise regression structural optimisation under small-sample constraints

This paper proposes a novel modelling approach, Mutual Information Clustering with Stepwise Regression, to address parameter drift caused by multicollinearity among hydrodynamic terms, which often occurs in model identification methods of ship manoeuvring that primarily depend on long-term training data. The proposed method initially clusters highly correlated hydrodynamic terms using mutual information and subsequently employs stepwise regression to identify those terms that significantly influence the output response. Validation is conducted utilizing the sea trial data from the full-scale YUKUN vessel, and the prediction performance is compared with the Abkowitz model, the manoeuvring model group model, and a conventional mutual information-based feature selection algorithm. The results indicate that the proposed algorithm effectively diminished hydrodynamic parameter drift, particularly when trained on small-sample datasets. • A novel modelling approach Mutual Information Clustering with Stepwise Regression is proposed, to address parameter drift. • The method initially clusters highly correlated hydrodynamic terms using mutual information. • Subsequently it employs stepwise regression to identify those terms that significantly influence the output response. • Validation is conducted using the sea trial data from the full-scale YUKUN vessel. • The proposed algorithm effectively diminished hydrodynamic parameter drift, even when trained on small-sample datasets.