Recursive Least Squares Identification For Output Nonlinear Systems Based On Decomposition

With the continuous development of control science and technology of system modeling, control object size becomes larger and larger, which makes the computational burden of system identification algorithm increase bigger and bigger accordingly. For the output nonlinear system whose structure is more complex than linear system, due to the product of parameter vector in the process of system identification, so that the information vector dimension increases, which leads to the sharp increase in the amount of calculation. It is imperative to explore the new parameter estimation algorithm whose calculation burden is small to identify output nonlinear system. Through the model decomposition and data filtering techniques, this paper studies the issue of recursive identification algorithm for output nonlinear system with colored noise. This graduation thesis focuses on “Recursive Least S quares Identification for Output Nonlinear Systems Based on Decomposition”and the proposed methods have profound significance in the aspect of improving computational e?ciency of nonlinear systems. The thesis is researched mainly from the following three directions.(1) For the output nonlinear equation error moving average(ON-EEMA) systems,output nonlinear equation error autoregressive(ON-EEAR) systems and output nonlinear equation error autoregressive moving average(ON-EEARMA) systems, based on the idea of over-parameterization model, this paper presented over-parameterization based recursive extended least squares algorithm、over-parameterization based recursive generalized least squares algorithm and over-parameterization based recursive generalized extended least squares algorithm.(2) In order to reduce the computational burden of the proposed algorithm, introducing the ideology of model decomposition. We decompose the ON-EEMA、ON-EEAR and ON-EEARMA systems into two virtual subsystems with smaller dimensions respectively, and propose the model decomposition based recursive extended least squares algorithm、model decomposition based recursive generalized least squares algorithm and model decomposition based recursive generalized extended least squares algorithm combined with recursive identification principle.(3) In order to further reduce the calculation amount, data filtering method is introduced on the basis of the model decomposition. For the ON-EEMA、ON-EEAR and ONEEARMA systems, data filtering based recursive extended least squares algorithm、data filtering based recursive generalized least squares algorithm and data filtering based recursive generalized extended least squares algorithm.To sum up, this paper mainly studies the recursive identification algorithm for output nonlinear systems based on model decomposition and demonstrates the e?ectiveness of the presented algorithm through the simulation experiment. Compared to the recursive least squares algorithm, the computational amount of the proposed algorithm is smaller but with similar parameter estimation accuracy.