Identification Modeling Of Ship Motions In Waves Based On Support Vector Regression

The motion prediction of ships in waves has been an issue of concern to the shipbuilding industry and is related to the navigation safety of ships.To predict the ship motions in waves,an important approach is to carry out numerical simulation with computer based on a mathematical model of ship motions.Traditionally,the methods for establishing the mathematical model of ship motions in waves mainly include model test,estimation by using empirical formulae,numerical computation based on the potential theory and computational fluid dynamics(CFD)method based on solving viscous flow.However,these methods are mainly applied to ship model,and some problems exist when using them for predicting the motions of a full-scale ship in waves,such as the scale effect in using model test results for full-scale prediction,and the limitation of the capability of CFD computation,etc.In the last decade,with the development of system identification technique and the constantly emerging of new identification methods,it has drawn an increasing attention from the shipbuilding industry to apply system identification technique in identification modeling and prediction of ship motions in waves.System identification technique,including parametric identification and nonparametric identification,is a method of determining the suitable mathematical model of a system by fitting the input and output of the system.The purpose of parametric identification is to estimate the unknown parameters in the mathematical model which has known structure and unknown parameters.The purpose of nonparametric identification is to establish a mathematical model for approximately describing the dynamic characteristics of a system,which has unknown mathematical model structure and parameters.Applying system identification technique can supplement the traditional prediction methods and enrich the technical means for modeling and prediction of ship motions in waves.Besides,system identification technique can be applied directly in the modeling and prediction of full-scale ship motions in waves,thus can avoid the influence of scale effect and improve the prediction accuracy.In this thesis,the support vector regression(SVR)is applied to the research on identification modeling of ship motions in waves.Compared to the identification methods based on conventional statistics theory,such as the least square method,artificial neural network,etc.,SVR is based on the modern statistic learning theory.It overcomes the shortcomings of the identification methods based on conventional statistics theory,such as worse generalization performance,easily trapped in the local optimum and only suitable for learning with large scale samples,and has better generalization performance,is especially suitable for learning with small scale samples.In order to improve the learning efficiency,the sequential minimum optimization algorithm is used to solve the convex quadratic programming problem of SVR,and the grid search method is used to determine the optimal penalty parameter and the insensitive loss factor of SVR.In order to verify the accuracy and validity of SVR in identification modeling of ship motions in waves,SVR is applied to the identification modeling of one degree of freedom(1-DOF)roll motion,coupled heave-pitch motion,and parametric roll resonance motion of ships in waves,respectively.To predict the roll motion of ships in waves accurately,the key issue is to determine the damping and restoring moments accurately.Because the roll damping is closely related to the fluid viscosity,so far a generally effective method for accurately predicting the roll damping of ships in waves is still absent.To this end,SVR is applied in this thesis to parametric identification and nonparametric identification of 1-DOF roll motion of ships in waves.In the aspect of parametric identification,SVR is applied to analyze the free roll decay motion in still water,the roll motion in regular waves and the random roll motion in irregular waves to identify the damping and restoring moment coefficients in the roll motion equation.Then the identified coefficients are utilized for numerical simulation of ship roll motion.When SVR is applied to analyze the ship roll motion in irregular waves,the random decrement technique is used to preprocess the random roll motion data to obtain the random decrement signatures of the roll motion.Based on the obtained random decrement signatures,the damping and restoring moment coefficients are identified by SVR.The identification results of the numerical simulation and the ship model test demonstrate that SVR can be applied to identify the free roll decay motion in still water,the roll motion in regular waves and the random roll motion in irregular waves to obtain the damping and restoring moment coefficients accurately.In the aspect of nonparametric identification,SVR is applied to analyze the free roll decay motion in still water,the roll motion in regular waves and the random roll motion in irregular waves to establish the nonlinear SVR models for the roll damping and restoring moments,and the identified results are utilized to simulate numerically the ship roll motion.In order to verify the identification effect of SVR,it is applied to analyze the numerical simulation data and the ship model test data,respectively.The identification results indicate that SVR can be applied to the nonparametric identification of the free roll decay motion in still water,the roll motion in regular waves and the random roll motion in irregular waves to establish the nonparametric models of the damping and restoring moments of the roll motion.The prediction of the coupled heave-pitch motion of ships in waves is a typical problem in the seakeeping investigation.In the past,the coupled heave-pitch motion is mostly predicted for ship model in waves by using numerical calculation method and model test,and few researches on motion modeling and prediction are carried out by using system identification technique.In this thesis,for the purpose of applying SVR to predict the coupled heave-pitch motion of full-scale ships in waves,SVR is applied to parametric identification and nonparametric identification of the coupled heave-pitch motion in regular and irregular waves,respectively.In the aspect of parametric identification,the hydrodynamic coefficients in the coupled motion equation are identified by using SVR.In the aspect of nonparametric identification,the SVR models of the hydrodynamic forces in the coupled heave-pitch motion are established.Considering that the motion of ships in irregular waves is random,the identification method based on a combination of random decrement technique and SVR is applied to identify the coupled heave-pitch motion in irregular waves.Therein,the random decrement technique is applied to obtain the random decrement signatures of the coupled heave-pitch motion,while SVR is applied to identification modeling and prediction of the coupled heave-pitch motion by analyzing the obtained random decrement signatures.In order to validate the identification effect,SVR is applied to analyze the numerical simulation data and the ship model test data of the coupled heave-pitch motion in regular and irregular waves.The identification results show that based on the obtained coupled motion responses,SVR can be applied to the parametric identification and nonparametric identification of the coupled heave-pitch motion in regular and irregular waves.When a surface ship,especially a container ship,navigates in the longitudinal waves,even though it sails in the moderate sea condition,the phenomena of parametric roll resonance may occur and seriously threaten the ship navigation safety.In order to predict the parametric roll motion accurately,in this thesis SVR is applied to the parametric and nonparametric identification modeling of the parametric roll motion of a container ship navigating in longitudinal regular waves.In the aspect of parametric identification,according to the known mathematical model structure of the parametric roll resonance,SVR is applied to identify the damping coefficients,restoring force/moment coefficients and wave exciting force/moment coefficients in the model.In the aspect of nonparametric identification,the SVR model for describing the parametric roll resonance of a ship in longitudinal regular waves is identified.In order to verify the identification effect of SVR,the parametric roll resonance of the container ship in longitudinal regular waves is numerically simulated,and SVR is applied to analyze the numerical simulation data to identify the unknown coefficients in the parametric roll mathematical model(parametric identification)or the unknown SVR model for the parametric roll resonance(nonparametric identification).The identification results demonstrate that SVR can be applied to parametric and nonparametric identification of the parametric roll resonance of a ship in longitudinal regular waves to establish the corresponding mathematical model.The SVR-based identification modeling method used in this thesis can not only be applied to parametric identification of ship motions in waves to obtain the unknown hydrodynamic coefficients in the mathematical model,but also can be applied to nonparametric identification of ship motions in waves to obtain the nonparametric SVR model and predict the ship motions in waves.It provides a novel and efficient method for the modeling and prediction of ship motions in waves.