ABSTRACT
Modern automatic voltage regulators (AVR) improve the terminal voltage responses, but they produce high gain and poor damping characteristics with low frequency oscillations 111 the system. To compensate these unwanted impacts of AVRs, the additiom11 signals generated by U1e device called power system stabilizer (PSS) are attached in AVR's feedback loop. The PSS, damps out the low frequency oscillation with auxiliary control signal provided to the excitation system, and improves the stability) of the electrical power system. The gain parameters of PSS are established on the origin of linearized model of electrical power system with a particular operating point, where they can suggest good arrangement. The design of conventional power system stabilizer (CPSS) based on the linearized model of power system (which is nonlinear in nature) cannot pledge its good performance in a practical operating atmosphere. The electrical power system being nonlinear and problematical subject comprises many features of circuit configurations, complex algebras, laws and other mathematical advances. This recommends and requires the PSS controller, which ought to posse.>s nature learning, adjustment aptitudes, handling the modifications and uncertainties of the system without having vast knowledge or identification of the system. The artificial neural network (ANN) possesses these all potential capabilities to deal with nonlinearity of the system, to
Model complex relationship ps and does not require the precise information of mathematical modeling or programming of system.
This thesis proposes feed forward neural network (FFNN) based PSS to improve tho:: performance and stability of electrical power system. The single machine at infinite bus (SMJB) system with AYR excitation and PSS is considered and its speed/frequency deviation and terminal voltage are taken as the inputs to radial basis function (RBF) and multilayer perception (MLP) architectures of FFNN. The proposed RJ3F-PSS with orthogonal least square (OLS) and MLP-PSS with back propagation (BP) algorithms are designed and compared with CPSS and PID-PSS. The simulation results of proposed PSS investigated in Matlab 7.13, Simulink 7.8 and Neural Network Toolbox 7.0.2, show the better performance, good settling time and less damping effects. The improvements of transient stability of terminal voltage
and dynamic stability of frequency deviations show the simplicity, suitability and better perfonnance of proposed technique.

(Xii)