Training an artificial neural network to discriminate between magnetizing inrush and internal faults

A feedforward neural network (FFNN) has been trained to discriminate between power transformer magnetizing inrush and fault currents. The training algorithm used was backpropagation, assuming initially a sigmoid transfer function for the network´s processing units (“neurons”). Once the network was trained the units´ transfer function was changed to hard limiters with thresholds equal to the biases obtained for the sigmoids during training. The off-line experimental results presented in this paper show that a FFNN may be considered as an alternative method to make the discrimination between inrush and fault currents in a digital relay implementation