Research On Subspace Theory Based Model Identification And Control Methods For Dynamic Positioning Ships

As an important tool in controller design and maneuvering forecast,how to establish a relatively precise mathematical model of a ship has been paid more and more attention.At present,traditional identification methods mostly focused on white-box modeling,i.e.constructing a mathematical model of a ship using empirical formular with ship type parameters,or through towering experiments to calculate related hydrodynamic parameters.These methods,however,suffered from complex calculation,or uneasy to carry out related experiments.In order to overcome these disadvantages,in this thesis we focus on black box model identification for dynamic positioning ships,and the corresponding control strategy.The discrete time system identification algorithm for ships is developed based on subspace theory,as well as subspace based continuous time identification algorithm.At the end of the thesis,the subspace theory is combined with model predictive control theory,giving a new controller called subspace model predictive controller,which can be used in the dynamic positioning system.And the content of the thesis is listed in the following in detail:As mentioned above,traditional methods focused on identifying hydrodynamic parameters of a ship through special self-propelled model tests,towering tests in a water tank,or empirical formulars.However,these methods suffered from high calculation complexicity,or parameters cancallation effects,more or less,thus leading larger errors in the identification results.In order to overcome these disadvantages,in this thesis we start at studying the identification problem for the nonlinear black-box ship model.In order to apply subspace identification theory,we first translate the ship model into the Hammerstein form with nonlinear feedback term according to the nonlinear characteristic of the traditional nonlinear differential equation of a ship.Then the nonlinear terms are expanded using Fourier series as base functions to simplify the identification procedure.Finally,according to the transformed model,we propose to apply MOESP algorithm to estimate the related parameter matrices,thus obtaining the black-box model of a ship.As some of the hydrodynamic parameters of a ship changes with diferent conditions of the ocean environment,in this thesis we propose an update strategy to online adjust the black-box model of the ship.The strategy is realized by using simplified Givnes rotation matrices to deal with the complex calculation in QR factorization according to its characteristic,with calculation complexicity reduced.Finally this strategy makes subspace identification online updated in the identification of black-box ship model identification.As we all know,different sampling frequency of input and output data will affect the precision of a model in the discrete time identification framework.So in this thesis,in order to overcome this problem,we propose to use continuous-time subspace identification algorithm to identification the continuous time state space ship model,which has advantages in controller design.In order to obtain high-order derivatives of the input and output data,we choose to use Laguerre filter functions and Laguerre projection functions to convert related signals,and finally translate these signals into“discrete time form”in corresponding function spaces.Then two kinds of continuous-time subspace identification algorithms,PBSID_filt and PBSID_opt,which are based on Laguerre filter functions and Laguerre projection functions,can be applied.The original system matrices can then be recovered by bilinear mappings between continuous time domain and the function spaces.Finally,in this thesis we propose the Laguerre filter based subspace predictive control algorithm?SMPC-L?by combining subspace theory,Laguerre filter function,and model predictive control theory.The algorithm,SMPC-L,overcomes the disvantage when traditional model predictive control usually estimate the model parameters at first.SMPC-L will predict the output of the ship in a single step,avoiding the procedure of estimating ship model parameters.What's more,by introducting discrete Laguerre filter functions,we can improve the efficiency of the algorithm by reducing variables in the optimization procedure,without affecting the control precision.