An efficient physics-data collaborative modeling of ship maneuvering motion based on Gaussian process regression and soft clustering strategy

An efficient physics-data collaborative modeling method for ship maneuvering motion based on Gaussian process regression (GPR) and soft clustering strategy is proposed in this paper. First, an integrated architecture that combines a physics-based dynamical model with a data-driven GPR residual dynamical model is designed. The physics-based model provides prior information on hydrodynamic forces and moments, while the GPR-based residual model learns the residual dynamics induced by unmodeled effects. Second, by introducing soft clustering based on a Gaussian Mixture Model (GMM), the computational efficiency of the collaborative model is improved. Two case studies involving the KCS container ship and the ONR Tumblehome ship are conducted, and comparative studies are performed among the proposed collaborative model, data-driven model, and physics-based model. The results demonstrate that the proposed efficient collaborative model has high prediction accuracy and computational efficiency, while retaining good physical interpretability.