Data Filtering Based Recursive Least Squares Identification For Output Error Models

It is well known that the fundamental problems in the control systems areacquired their mathematical models, and system identification technique is anapproach of establishing mathematical models of systems. The research results ofsystem identification is widely used in scinece and engineering. Therefore, systemidentification techique has been an active research area for the last few decades. Inthis dissertation, we focused on three different kinds of output error models withtime-variable parameters. All of the algorithms are based on the method of datafiltering and auxiliary model recurise square identification technique. The maincontents of this dissertation are as follows.For output error moving average model, the forgetting factor is introduced tothe method of auxiliary model based recurise square identification technique, andfor a special class of output error moving average model with time-varialbeparameters, we come up with a kind of recurise square identification method whichbased on data filtering. This algorithm reduce the estimated error that caused by theforgetting factors.For a kind of output error autoregressive model which contains time-varialbeparameters, using a linear filter to divide it into system model and noise model, andaccording the method of auxiliary model to design the algorithm for the filteredmodels. This method reduce the dimension of the model, and the biggest advantageis that we can design forgetting factor for each system individually. The algorithmcan be extensively and flexibly used.Based on the idea of data filter, auxiliary model and the technique of designingforgetting factor individually, we proposed a recursive square algorithm for theoutput error autoregressive moving average model which contains time-varialbeparameters, and verified the effect of parameters estimation by simulation.