الفهرس 
Literature ReviewOnly 14 pages are availabe for public view
 |  | from | 16 |  |  |
| | | from | 16 | | |
Abstract
Most ”real world” systems relevant from a control system perspective aie nonlinear behavior and furthermore they are often hard or unrealistic to derive their model using the well known laws of physics. The conventional controller design for such systems is usually difficult to derive and of complex realizations
The artificial neural netwoiks (ANN) provide a very promising tool to build good engineering system models from their measured data alone. This is mainly due to the ability of ANN to act as a function approximator with a high degree of accuracy
If a neural network system model is available, then there exists dif’feient approaches for the design of controllers, e.g., feedback linearization, or linearization of the model followed by a linear design There is the other approach of using ANN as a system controller A one widely used scheme for the use of ANN in controlling nonlinear plants is to identify the inverse dynamics of the unknown plant and then apply an inverse identification model as part of the feedforward controller
In this thesis, different ANN topologies are used, namely the feedtorwaid neural network (FNN) with a generalized weight adaptation algorithm, and the diagonal recurrent neural network (DRNN) with a generalized dynamic back-propagation algorithm for the forward and inverse identification of nonlinear dynamic systems. Then a study is made of neuro-based controllers using different topologies of ANN, with both on-line and off-line training methodologies to investigate their effects on controlled system peiiomiance The study is based on controlling the plant using the idea of what is called a coordination of a feedforward controllers combined with an inverse system dynamics identification mode The feedforward controller generates the control actions to keep the plant output at a given reference value at the steady state, while other controller generates the control actions to stabilize the system transient tracking error-Simulation examples are used to study the effects of varying the W\ topology and its training methods of the neuro-controllers on the overall controlled system pa toimance.