Advanced controller designs for synchronous generators using nonlinear transformations and neural networks.

Abstract

The non-availability of a technique to handle all nonlinear problems necessitates a multitude of techniques be attempted to facilitate the solution for a particular problem. This dissertation investigates the development of advanced design techniques for nonlinear controllers and observers for a synchronous generator. The primary objective of the control system in this application is to asymptotically track the nominal terminal voltage and frequency, and to improve the stability and dynamic performance in the presence of both small and large disturbances. New schemes for designing nonlinear controllers and observers for the synchronous generator are introduced using the concepts and methods from differential geometry and neural network theory. Nonlinear transformation using differential geometry methods is a promising approach to the control of nonlinear systems. The differential geometry-based nonlinear controllers are developed for the synchronous generator and these control schemes are tested on different order models. The initial controller design is conducted based on a nominal load consideration. Later, on-line stabilization which consists of updating the controller parameters is conducted to compensate for the load variations. Differential geometry concepts are use to develop a new nonlinear observer for a synchronous generator that has the ability to handle large transients. The investigation conducted here also includes other kinds of nonlinear observers and compares them with their linear counterparts for different fault clearing times. A software program using symbolic manipulations is developed to help the user, who may not be familiar with differential geometry, to design a nonlinear observer for both single and multi-output models of the synchronous generator. Neural networks provide a solution to interesting identification and control problems. Their parallel nature assures a fast adaptation to the system under control. A new control scheme is developed to adaptively control the frequency and voltage of a synchronous generator using neural networks. Dynamic neural networks whose parameters are identified via a supervised learning procedure are used as approximators to the nonlinear map given by the system input-output data. The trained neural network is then used for the design of the adaptive control scheme. The performance of these newly developed controller and observer schemes are examined in detail for several different cases. It is shown that the presently designed controllers and the observer provide an improved performance when compared with the existing design procedures.