A neural optimal voltage regulator

Two desirable features of artificial neural networks (NNs) are that they can implement parallel processing and can learn nonlinear functions. This paper reports a NN application that makes use of these two important features. A generator excitation control system is a nonlinear system. The conventional way to design an optimal controller for this system is to linearize the system at several selected operating points and implement optimal control at these points separately. In this paper, a neural network is trained to give the optimal control gains over the whole operating range of the excitation system. The input of the NN is the power angle of the generator and the outputs are the optimal control gains. The inherent parallel processing feature makes this design applicable for online applications.