Neural Network Regulators for Synchronous Machines

Abstract Much research has been conducted in an effort to improve dynamic power system performance through modulation of the excitation system. Power system stabilizers are commonly used to provide the required regulation. Since a power system stabilizer is typically designed using a linearized system model, the regulation is applicable to a narrow system operating range. As a result, the stabilizer becomes less applicable for synchronous machines that have a wide range of operating conditions. The regulator presented in this paper combines neural networks with a self-tuning regulation strategy. The neural network is pretrained to approximate the inverse dynamics of the synchronous machine operating at a nominal operating point. It is then implemented as the controller for the system and allowed to continue its training online. The operating point of the system can then be varied, and the neural network will adjust to the new operating point.