Two-Stage Identification Methods Based On Data Filtering

Based on the National Nature Science Foundation of China (NO.60973043), this thesis studies the two-stage identification methods for linear and nonlinear systems with colored noises. After reading some relevant references, the author briefly reviews the history of system iden-tification and overviews the existed parameter estimation methods in the exordium, and then derives the two-stage identification methods for different system models in detail in the next chapters. The main results are as follows:1. For the output error type systems with colored noises, the two-stage identification methods are developed. The main idea is to design a linear filter according to the structure of noise model, and to transform the system model into an output error model with white noise by filtering the inputs and outputs of the system by the corresponding filter designed, and then to interactively identify the transformed system model and noise model by using the auxiliary model identification idea and the least squares principle. The simulation results show that the proposed algorithms are effective.2. Many nonlinear systems can be described by Hammerstein systems, the two-stage iden-tification idea is extended to the Hammerstein dynamic adjustment model. Using the polynomial C(z) to filter the nonlinear inputs and outputs, the system model is trans-formed into a Hammerstein controlled auto regressive model. Then using the least squares principle to interactively identify the transformed system model and the noise model, the two-stage identification method based on data filtering is developed. The simulation exam-ple tests the proposed algorithm and compares with the recursive generalized least squares identification method and the stochastic gradient identification method.3. For the Hammerstein nonlinear systems disturbed by the general colored noise, i.e., the noise with a auto regressive moving average (ARMA) model, the two-stage identification method based on data filtering is proposed. Based on the data filtering and the least squares principle, that is, the identification process contains two steps, the system model and noise model identification. Simulation shows that proposed algorithm can give high accurate parameter estimates.4. For nonlinear output error type systems, the two-stage identification method based on data filtering is derived, combined with the advantages of auxiliary model algorithms and data filtering. The unmeasurable outputs and the unknown noise terms in the information vector are replaced with the outputs of an auxiliary model and estimates residual, respectively. Simulation examples demonstrate the method works well.In summary, this thesis proposes the two-stage identification algorithms for output error linear systems and Hammerstein nonlinear systems, the performances of the algorithms are illustrated by computer simulations. The convergence of the two-stage identification algorithms need further proof.