الفهرس 
1. INTRODUCTIONOnly 14 pages are availabe for public view
 |  | from | 110 |  |  |
| | | from | 110 | | |
Abstract
The requirements for the quality of automatic control in the process industries increased significantly with the increased complexity of the industrial processes and sharper specifications of product quality. The majority of industrial processes contain strong nonlinear relations, time-varying characteristics and high degree of uncertainties. Accurate models are required for simulating the process behavior, also for controller design relying on these models.
During the last two decades, much success has been achieved in the use of neural networks for modeling and control. It has been shown that neural networks are effective soft-computing techniques to model and control a broad category of complex nonlinear systems, especially to those systems whose mathematical models are extremely difficult to obtain. In general, the neural networks for the identification of dynamic systems fall into static neural networks (SNNs) and dynamic neural networks (DNNs). A dynamic neural network is a nonlinear dynamic system described by a set of nonlinear differential or difference equations with extensive connection weights. Hence, this type of neural networks is more suitable to be represented in state space format. The use of the state space representation, which is quite commonly used in most control algorithms, presents several advantages that must be taken into account.
In this thesis the DNN is used to model the dynamics of unknown nonlinear systems. Two DNN structures are used to model the nonlinear systems. Also an indirect adaptive controller based on DNN is developed in the form of the internal model control structure. This control method includes two learning phases, i.e., off-line and on-line learning. In the off-line learning, the DNN is learned by the epoch wise back propagation through time (BPTT) method to represent the forward dynamics of the system to be controlled. In the online phase, the DNN is used as the internal model of the controlled system and its parameters can be trained by the Truncated BPTT to cope with the possible change in the system dynamics. Hence, the mathematical inversion of the DNN internal model is computed on-line to act as the forward controller. Finally, the controller is then obtained by cascading this inverse model with a robust filter and a linear compensator to improve the closed loop performance.
In this thesis also, we develop a direct adaptive control scheme based on DNN for controlling of nonlinear systems. The DNN is represented in a general nonlinear state space form for producing the control action that forces the system output to a desired trajectory. The control algorithm can be implemented without a priori knowledge of the controlled system. Indeed, the weights of the DNN controller are adjusted on-line using the truncated BPTT method. Unlike the approaches in the literature, the learning signal of the network weights is generated by a control error estimator stage in the developed controller. Finally, the developed controllers are applied to a continuous stirred tank reactor (CSTR) with time varying behavior to evaluate the performance of controllers. The performance of controllers also evaluated by a laboratory flow control benchmark system with variable set-point and with adding an external disturbance.