Adaptive excitation and governor control of synchronous generators using multilayer recurrent neural networks

A novel neural network structure for the design of an adaptive control strategy for a single synchronous generator unit connected to a large power system through a transformer and transmission lines is presented. Both excitation control and governor control mechanisms are developed by exploiting the input-output mapping capability of trained neural networks for identifying the nonlinear system dynamics. A multilayer network architecture with a hidden layer that permits recurrent connections is used together with an LMS (least mean square) updating rule for supervised training to realize superior performance features in the excitation control and the governor control schemes