Online non-parametric 4 DOF modelling of ship motion using hybrid kernel relevance vector machine

The complex characteristics of ship motion under various manoeuvres and environmental conditions are determined online by a non-parametric identification model. This paper introduces a novel online ship motion modelling method that adaptively adjusts the training samples based on a sparse hybrid kernel relevance vector machine. Firstly, to fully capture the motion characteristics of the ship, the manoeuvrability test data is collected from a marine manoeuvring simulator, including different operators, engine orders, rudder angles and environments. Additionally, full-scale sea trial data is obtained from the YUKUN vessel, a ship type different from the virtual ship in the simulator. Subsequently, an adaptive criterion for updating the training sample is developed, and a sliding time window is applied to dynamically update the ship motion data. Finally, the motion states and trajectories of both ships are predicted under different conditions. These results show that the proposed method is well-suited for online non-parametric modelling of different ships and manoeuvre conditions. The proposed scheme's mean squared error and mean absolute error are below 0.10 and 0.16. Compared with the non-adaptive online scheme, the proposed approach has stronger generalisation and adaptability. This scheme can serve as a reference for online non-parametric modelling of unmanned surface vehicles. Various manoeuvres and environmental conditions are determined online by a non-parametric identification model An online manoeuvring method adaptively adjusts the training samples based on a sparse hybrid kernel relevance vector machine Maneuverability test data is collected from a maneuvering simulator and full-scale sea trial data is obtained from YUKUN An adaptive criterion updates the training sample, and a sliding time window is applied to dynamically update the data The motion states and trajectories of both ships are predicted under different conditions with high accuracy