Studies on a multi-machine power system with a neural network based excitation controller

A feed forward neural network based generating unit excitation controller, its implementation and experimental studies on a single-machine infinite bus system have been reported earlier. The proposed excitation controller integrates the AVR and PSS functions. It is trained on-line from the inputs and outputs of the generator on which it is installed. The error function used for training the neural network by the back propagation algorithm is modified by supplementary signals (speed and acceleration) to provide the stabilizing function. The proposed algorithm has fast convergence rate and high accuracy during on-line training. The performance of the proposed excitation controller to damp multimodal oscillations has been investigated on a five-machine power system without infinite bus. The generating units are connected through a transmission network to form two areas. The model multimachine system configuration is such that it exhibits multi-modal oscillations when subjected to a disturbance. Simulation studies described in the paper show the effectiveness of the proposed integrated controller. It provides a good AVR function of effective terminal voltage control and also damps both the local and inter-area modes of oscillations very effectively