An Implementation of a Continually Online Trained Artificial Neural Network Controller for a Turbogenerator

(A full thesis option MSc(Eng) topic at University of Natal Durban South Africa, completed in Decemeber 1998)

by

Ganesh Kumar Venayagamoorthy

Abstract

Turbogenerators supply most of the electrical energy produced by mankind and directly affect the security and stability of electrical power systems.  A turbogenerator is a nonlinear, fast-acting, multivariable time variant system, and is usually connected through a transmission system to the rest of the power system.  Conventional automatic voltage regulators and turbine governors are designed to control, in some optimal fashion, the turbogenerator around one operating condition; at any other operating condition the generator's performance is degraded

Continually Online Trained (COT) artificial neural networks (ANNs) are able to identify nonlinear processes such as turbogenerators and to control them in some desired fashion.  This thesis investigates the practical implementation of a COT ANN based regulator for a laboratory turbogenerator system.  The regulator consists of two separate ANNs. The first ANN is an intelligent identifier and the second is an intelligent controller.

Mathematical models of the laboratory turbogenerator and the two ANNs are derived and modelled in MATLAB and SIMULINK and are used to show in simulation that the ANN identifier can identify the turbogenerator dynamics accurately and the ANN controller can control the turbogenerator under dynamic and transient conditions as well as the conventional automatic voltage regulator and governor.  When system conditions change such as different power levels, and transmission line configurations, the regulator system tracks these changes and its performance does not degrade as in the case of the conventional automatic voltage regulator and governor system.

The ANN system was practically implemented on a single Intel 486 microprocessor platform in the “micro-machine” research laboratory at the University of Natal, Durban.  Measured results agree well with predictions, and show that ANN controllers have potential to control large turbogenerators in power stations.  However, much work still remains before these ideas could be implemented on such large machines.  Because they give an improved dynamic response, they allow the turbogenerator to operate more closely to its steady state stability margin and thus produce more electrical power per invested Dollar of plant.