Evolutionary improvement of neural classifiers for generator out-of-step protection

In the paper the results of investigation are presented on application of artificial neural networks to protection of synchronous machines against out-of-step conditions. The new ANN-based protection ensures faster and more secure detection of generator loss of synchronism when compared to traditional solutions. To determine the most suitable network topology for the recognition task a genetic algorithm is proposed. The final ANN structure is found using the rules of evolutionary improvement of the characteristics of individuals by concurrence and heredity. The proposed genetic optimisation principles have been implemented in Matlab programming code. The initial as well as further consecutive network populations were created, trained and graded in a closed loop until the selection criterion was fulfilled. The results of thorough testing of the designed protection scheme with ATP-generated power system signals are described and discussed.