Research On The Parameter Identification Methods Of Hammerstein-Wiener Multivariable System

The block structured multivariable nonlinear systems composed of the static nonlinear link and dynamic linear link widely exist in the practical production processes,such as electrode regulating system of electric arc furnace,Mach number control system of wind tunnel flow field,chickness control system of rolling mill,blowing prediction system of P-S copper converter,etc.Although the physical structure of this kind of system is clear,the signals between the links are unknown and there are also multivariable coupling in the sytem.The results of parameter identification of this kind of system are few and imperfect.Therefore,to study the parameter identification of multivariable nonlinear systems with block structure is of great theoretical significance and practical value.In this dissertation,the parameter identification method of H-W(Hammerstein-Wiener)multivariable nonlinear system is studied,which consists of input nonlinearity link,intermediate linearity link and output nonlinearity link.The main research work is as follows.(1)For the H-W multivariable nonlinear systems with coupling characteristics only in the output nonlinearity link,a parameter identification method based on separable nonlinear least squares(SNLS)algorithm is proposed.The problem of computing generalized inverse of matrices and projection operators is solved by matrix decomposition operation.The convergence of proposed SNLS identification method is analyzed by the maximum posteriori estimation principle.The validity of proposed SNLS identification method is verified by the computer simulation.Furthermore,the electrode regulating system of steelmaking electric arc furnace can be described by this type of multivariable nonlinear system.Using the actual data collected in the field,the parameters of the electrode regulating system are identified successfully by the method presented in this dissertation.It lays a foundation for the output prediction and controller design of the electrode regulating system.(2)For the H-W multivariable nonlinear systems with coupling characteristics in the intermediate linearity link and the output nonlinearity link,a parameter identification method based on interior point method(IPM)and Levengerg-Marquardt(LM)algorithm is proposed.The convergence of the proposed IPM-LM identification method is analyzed by the convexity of the augmented identification criterion and the Armijio step search method in LM algorithm.The effectiveness of the proposed IPM-LM identification method is verified by the computer simulation.(3)For the H-W multivariable nonlinear systems with coupling characteristics in the input nonlinearity link and the intermediate linearity link,a parameter identification method based on recursive auxiliary variable least squares(RAVLS)algorithm is proposed.Choosing the auxiliary variables,the estimation bias of multivariable least squares operation is solved,and the recursive identification of model parameters is realized.The uniform convergence of the proposed RAVLS identification method is proved according to martingale convergence theorem and multivariable least squares unbiased estimation theory.The validity of the proposed RAVLS identification method is verified by the computer simulation.(4)For the H-W multivariable nonlinear systems with coupling characteristics in the input nonlinearity link,the intermediate linearity link and the output nonlinearity link,a parameter identification method based on the genetic ant colony algorithm(GACA)and the Kalman filter(KF)algorithm is proposed with the identification criteria,which is the minimum sum of squares of deviations between model output and actual output.The convergence of the proposed GACA-KF identification method is analyzed by the finite homogeneous Markov process characteristics of the optimal solution sequence and the non-negative upper martingale property theory of the identification criterion value sequence.The validity of the proposed GACA-KF identification method is verified by the computer simulation.