Implementation and laboratory test of a fully integrated neural network based protection relay for high voltage transmission lines

This paper presents a novel approach to fault detection, faulted phase selection and direction estimation based on artificial neural networks (ANNs). The main target in this work is to find a relay design that is fast with a detection time independent of fault conditions (i.e. current transformer saturation, dynamic arcing faults, short circuit level and system topology) and that uses only unfiltered voltage and current samples, with the same structure and parameters for each relay location in a network. The neural relay is trained globally using training patterns from more than one relaying position in order to be as generic as possible. The CAD of the ANN-based relay is implemented into a hardware model (a DSP evaluation module plugged into a PC) and tested under laboratory conditions via the real-time digital simulation (RTDS) system.