Research On Ship Motion Modeling Method Based On Gaussian Process Regression

With the acceleration of the development of ship intelligence,autonomy,and unmanned,the research on the testing technology of intelligent navigation ability for ships has attracted extensive attention of scholars at home and abroad.Virtual simulation testing has the advantages of low cost,zero risk,customization,acceleration,and repeatability.However,the accuracy of the simulation is questionable and there is a lack of real feedback.Physical experiments can get real testing feedback,but it is limited by the safety,economy,and efficiency of testing,and the controllability and repeatability of testing are poor.Therefore,it is urgent to carry out the research of testing technology for intelligent navigation of ships.Ship motion control in virtual simulation and physical experiments depends on an accurate motion model.Dynamic system modeling methods can be divided into mechanism analysis modeling and datadriven modeling,and data-driven modeling can be subdivided into parametric modeling and nonparametric modeling.In the field of ship motion modeling,the research on mechanism modeling and classical parameter modeling methods has been relatively mature.With the development of modern engineering automation,machine learning has gradually become a popular research method.Learning nonparametric modeling methods are more and more used in the field of ship motion modeling.This paper analyzes the testing technology of intelligent navigation for ships,introduces the theory of ship motion modeling and the classical ship motion modeling methods,and studies the ship motion modeling method based on Gaussian process regression(GPR).The simulation testing is carried out by using the mechanism model constructed by the experiment data of container KCS and tanker KVLCC2 provided by the ITTC.The model-scale testing is carried out based on the container ship and harbor tug on the model-scale ship experiment platform,and the NNs,LS-SVM,and GPR are used for motion modeling and comparison.The main contents of this paper are as follows:1)Testing technology of intelligent navigation for ships.This paper analyzes the requirements of intelligent navigation testing for ships and puts forward a testing method of intelligent navigation for ships with virtual testing as the guide,model-scale experiment as the pilot testing,and full-scale experiment as the verification.Then combs the testing contents of intelligent navigation for ships,and finally points out the relationship between intelligent navigation test and motion modeling.2)Theory and method of ship motion modeling.The ship motion coordinate system and coordinate system transformation are introduced;The Abkowitz model,MMG model,and Nomoto model are introduced,and the three models are compared;This paper introduces the theory of ship hydrodynamic maneuvering,introduces two classical ship motion modeling methods(NNs and LS-SVM)and the identification principle in this paper;Finally,the principle of GPR based on function space,the covariance function of GPR and the super parameter optimization method are introduced.3)GPR motion modeling and simulation verification.This paper introduces the overall framework of ship motion modeling based on GPR and uses NNs,LS-SVM,and GPR to model and compare the simulation experiment data generated by the mechanical models of container KCS and tanker KVLCC2.The results verify the feasibility and effectiveness of the GPR modeling method to support the model-scale ship verification.4)GPR motion modeling and experimental verification.The model ship,hardware equipment,and software system of the model-scale ship experiment platform are introduced;The model-scale container ship and harbor tug are used to carry out turning experiments and zigzag experiments to obtain data.The NNs,LS-SVM,and GPR are used to model and compare experiment data,which verifies the effectiveness and universality of the GPR modeling method for the model-scale ships.