الفهرس Only 14 pages are availabe for public view
 |  | from | 122 |  |  |
| | | from | 122 | | |
Abstract
Identification of linear and bilinear systems is the main interest of this thesis system identification, which is the field of mathematical modeling of systems from perimental data, has acquired widespread applications in many areas. These elude automatic control, spectral analysis, pattern recognition, simulation, aptive filtering and linear prediction. The class of bilinear systems (BLS) ceives a remarkable attention in the last few decades. The current interest in BLS as due to its closeness to linear systems in which the familiar analytic tools for near system can be used to advantage in BLS. On the otherhand, approximating a onlinear system by a bilinear model results in better accuracies than approxima-ng them by a linear model. This conclusion has hen verified in the present work. different identification models are considered, i.e. autoregressive with.ogenous variables (ARX) , autoregressive ill’ -ving average with exogenous ariables (ARMAX), and output error (OE) model structures. Estimation of these odes ’parameters’ is based on the recursive leas. squares algorithm. One main objective of this work is to obtain a faster convergence rate of the gorithm. In this thesis, a new stopping criterion is suggested and a remarkable provement in the convergence rate is obtained. plementation of these algorithms for different numerical examples prove that work well in practice compared with other algorithms already presented in apter 2. order to determine the best model structure, a successful use of the so-called Oung’s information criterion (YIe) is proved. An additional main and important objective of the present work is to realize ( ascmatmg feature. of the BLS structure. For this purpose, an exact nonlinear.