Neural network identification of power system transfer functions

Abstract

This paper describes an investigation into the use of a multilayered neural network for measuring the transfer function of a power system for use in power system stabilizer (PSS) tuning and assessing PSS damping. The objectives are to quickly and accurately measure the transfer function relating the electric power output to the AVR PSS reference voltage input of a system with the plant operating under normal conditions. In addition, the excitation signal used in the identification procedure is such that it will not adversely affect the terminal voltage or the system frequency. This research emphasized the development of a neural network that is easily trained and robust to changing system conditions. Performance studies of the trained neural network are described. Simulation studies suggest the practical feasibility of the algorithm as a stand-alone identification package and as a portion of a self-tuning algorithm requiring identification in the strategy. The same technique applied to a forward modelling scheme can be used to test the damping contribution from different control strategies.