Multi-Innovation Identification Methods For Input Nonlinear Systems

The identification for nonlinear systems is difficult and has been a hot topic in the field of control theory.As a typical block-oriented model,the input nonlinear systems provide a good compromise between the tractability of linear systems and the accuracy of nonlinear systems,and thus have been widely used in various fields.The basic idea of multi-innovation identification theory is to extend the dimension of the innovation vector,thus increasing the utilization rate of observation data and improving the algorithm performance.The thesis considers the identification problems for input nonlinear systems by combining the multi-innovation identification theory with new technologies and ideas in the field of system identification.The major results are as follows.(1)For the input nonlinear equation error systems,a over-parameterization based multiinnovation(MI)gradient identification algorithm is derived by means of the overparameterization method.To reduce the effect of colored noise on the parameter estimation and acquire more accurate parameter estimates,a linear filter is used to filter the input-output data and “whiten” the colored noise,and a over-parameterization based filtering MI gradient identification algorithm is proposed.Furthermore,the proposed algorithms are extended to identify the multivariable input nonlinear equation error systems.(2)For the input nonlinear equation error systems,to solve the problem of the redundant parameters caused by the over-parameterization method,the key-term separation method is applied to parameterize the models,and a identification model which does not contain the redundant parameters is obtained.Then a key-term separation based MI gradient identification algorithm is presented by using the data filtering technique.Moreover,a new parameterization method which called the parameter separation method is proposed,its basic idea is to use a linear filter to release the multiplicative relationship between the parameters of the linear and nonlinear blocks,and thus all the parameters to be estimated are separated and appear explicitly in the model description.Based on this method,some parameter separation based MI gradient identification algorithms are proposed.(3)For the input nonlinear output error systems,based on its bilinear-in-parameter model,by constructing two augmented information matrices,the system is decomposed into two submodels.Then according to the auxiliary model identification idea and hierarchical identification principle,the unknown variables is replaced by the outputs of the auxiliary model and the parameters of the two submodels are estimated interactively,then a bilinear-in-parameter model decomposition based auxiliary model MI gradient identification algorithm is proposed.The method can avoid estimating the redundant parameters and has higher computational efficiency.(4)For the multivariable input nonlinear output error systems,based on the key-term separation method and the Kronecker product operation,a key-term separation based auxiliary model multivariable MI gradient identification algorithm is presented.For the purpose of avoiding generating information matrices with large dimensions and computationally intensive problems caused by the Kronecker product,the model is decomposed into several submodels based on the parameter types(parameter vectors or matrices),and a key-term separation based decomposition auxiliary model multivariable MI gradient identification algorithm is proposed.In the thesis,the simulation examples are provided to illustrate the effectiveness of each proposed algorithm.The computational burden of some algorithms are analyzed and compared.