System identification and control using multivarate statistical methods

The main focus of this thesis is to exploit the ability of the multivariate statistical methods which yield reliable solutions to many industrial problems where the ordinary regression methods tend to fail. The principal components analysis (PCA), principal component regression (PCR) and partial least squares (PLS) are some of the very popular MVS technique. Majority of the applications of these tools in process industries are particularly seen in process monitoring, identification, estimation and controls. Traditional modeling techniques like ordinary least squares (OLS) are very sensitive to the quality of the data. The estimates derived from these techniques are generally poor if the data is not sufficiently rich. Majority of the chemical processes are time varying in nature and the traditional online adaptation methods use LS based algorithm to update the time varying parameters. However, these techniques have been found to suffer from large variance errors when the past data is discounted continuously and if the newer data brings no information regarding the current state.


source :

Tags :
Your rating: None Average: 5 (2 votes)