An adaptive neurocontroller for speed control of a synchronous generator

Abstract

Neural networks have shown great promise in many areas of engineering. In this paper, we present a newly designed neural control system that consists of three neural networks cascaded together, one representing the inverse model of the speed-governing and turbine system, another identifying the dynamics of the synchronous generator, and a third being part of the controller. The inverse model is achieved with a multilayer feedforward neural network trained in batch mode through back-propagation learning. Once the network is trained, its weights and biases will be fixed. The dynamics of the synchronous generator is identified on-line while the generator is operating. The weights of the neurocontroller are determined by sweeping back the control error. Usually this updating process has a lower frequency than the identification process to ensure the stability of the entire control system. The neurocontroller was applied to a multi-machine power system and some simulated results are presented.