Research On Identification Methods For Systems With Colored Noises

System identification theory is a methodology describing various system dynamic characters. It is an efficient tool for system studying, using the tool we can depict the system quantitatively. From the sixties last century, as the rapid development of modern control theory, system identification had been going development. Recently, system identification theory has been successfully applied in some engineering fields. But there are still some difficulties to figure out. The thesis study the identification of systems with colored noises, which totally meet the need of the engineering application, expanding application, expanding some new identification methods.The author research and read lots of related references in the literature and study identification for stochastic systems with colored noises. The main contribution are as follows.1.The thesis presents a multi-innovation stochastic gradient method based on model equivalence principle for General system, one of the system with colored noises. The basic idea was to approximate the process model and noise model by using FIR models, and then to obtain a CARMA model which by multi-innovation stochastic gradient algorithm, and finally to determine the parameters of the original systems by means of the model equivalence. The proposed algorithm requires less computational efforts and can enhance the parameter estimation accuracy by enlarging the innovation length in simulations.2.For multi-input, multi-output stochastic systems with correlated noises, this paper proposed an auxiliary models(FIR) identification method of multivariable systems. The basic idea was to approximate sub-submodel of transfer matrices by means of auxiliary models--- finite impulse response(FIR) models, and then to obtain auxiliary models to estimate the outputs of the sub-submodel, and further to use the recursive least squares algorithm or the Pade approximation method to produce the parameter estimates of sub-submodels, the simulation results confirms the theoretic findings.