Research On Parameters Identification Algorithms For Hammerstein-Wiener Nonlinear System

The mechanistic models of practical complex industrial control systems usually cannot be derived from mathematical,physical or other theorems.This kind of systems should be modeled by data.In other words,system models can be established by employing system identification methods.System identification has been widely implemented in many areas such as signal processing,fault detection and diagnosis,adaptive control.In fact,well developed linear identification methods cannot be applied since some systems possessing strong nonlinear characteristics are unable to be linearized around their operating points.Therefore,it is significant to investigate algorithms which are suitable for the identification of nonlinear systems.Currently,a class of block-oriented nonlinear structures has become the hotspot in the research on parameter identification of nonlinear systems.A wide range of practical industrial processes with nonlinear characteristics can be described by using Hammerstein-Wiener nonlinear systems effectively.Therefore,this thesis chooses the Hammerstein-Wiener nonlinear system as the control object.Two system identification algorithms are proposed in this thesis to estimate parameters of the Hammerstein-Wiener nonlinear system.The major work of this thesis can be summarized as follows:(1)For the single-variable Hammerstein-Wiener nonlinear system,an auxiliary model is introduced in order to solve the problem that some variables in the identification model cannot be measured.Then,the hierarchical least squares algorithm which needs less computational effort comparing to other algorithms is employed in order to identify parameters of the system.(2)According to the limitation of above algorithm,only the Hammerstein-Wiener system with reversible nonlinear output element can be identified,so a single-input single-output Hammerstein-Wiener neural network is designed.In this case,the problem on parameter identification of system is transformed into the weight adjustment of the network.Furthermore,the momentum steepest descent method is used to train the network.Meanwhile,in this way,the low accuracy phenomena which could be caused by the auxiliary model can be avoided.(3)Aiming to identify the multivariable Hammerstein-Wiener nonlinear system that has more parameters and more complicated structure,there are two' cases to study.Firstly,a hierarchical least.squares algorithm and a momentum steepest descent method are exploited respectively to identify parameters of the MIMO Hammerstein-Wiener system with coupling condition only in the linear dynamic link.Secondly,a momentum steepest descent method is used to identify parameters of the MIMO Hammerstein-Wiener system with coupling condition both in the input nonlinear element and the linear portion.(4)The convergence,validity and accuracy of two proposed algorithms are verified by using computer simulations.