ARTIFICIAL NEURAL-NETWORK-BASED METHOD FOR TRANSIENT-RESPONSE PREDICTION

The paper describes the application of an ANN-based approach to the pr ediction of the dynamic behaviour of a synchronous generator following a disturbance in a simple power system. Case data representing the os cillation in rotor angle caused by the disturbance is accumulated from offline simulation using an accurate digital model. Neeural networks are trained to map this case data in relation to the initial operating conditions and details of the particular disturbance involved. The re sponse of a one-machine, infinite-bus system is considered after the o ccurrence of a three-phase short circuit in one element of a double tr ansmission line connecting the synchronous generator to the bus, and o f a shock local load change. Numerical results comparing the predicted response using the ANN model with that obtained from direct use of th e benchmark model are presented in terms of both accuracy and speed. T hese suggest that the ANN model might be used in conjunction with an o nline fault identification system in the study of transient stability or in the provision of information for predictive control. Because onl y very simple mathematical calculations are required once the neural n etworks have been trained, the computation time is very short in compa rison with the direct use of numerical simulation.