Neural networks using apparent resistance for out-of-step relaying

This paper describes methods used for training neural networks (NNs) to perform out-of-step relaying using apparent resistance and its rate of change. NNs have an adjustable output threshold value that changes the numbers of failures to trip and false trips. The performance of NNs for out-of-step relaying is not always good over a wide range of output threshold values. The method described in this paper called the confusional filtering criterion (CFC) helps produce lower error rates for a wider range of output threshold values. This method modifies the training set before training the NN. The training set contains input-output pairs consisting of R and Rdot measurements plus the desired output, which is whether the relay should or should not trip.