Neural network-based algorithm for power transformer differential relays

Abstract

A neural network-based algorithm for the protection of a one-phase power transformer is considered. The neural network input is a four-dimensional vector obtained by a fast-frequency analysis of the differential current. Primary and secondary currents are measured with current transformers the saturation of which is taken into account. A set of training cases is generated with the help of the electromagnetic transients program. The neural network-based algorithm is compared with a conventional differential algorithm. It is found to be more efficient, especially in the case of saturation of the current transformers.