A New Approach to Phase Selection Using Fault Generated High Frequency Noise and Neural Networks

Single-pole autoreclosure is quite extensively used in long-line applications and involves tripping only the faulted phase for single-phase earth faults. Reliable and fast phase selection is thus imperative in order to avoid potential problems of system insecurity and instability. Conventional phase selectors, primarily based on power frequency measurands, can suffer some impairment in performance because of their heavy dependency on varying system and fault conditions. However, the advent of artificial neural networks (ANNs), with their ability to map complex and highly nonlinear input/output patterns, provides an attractive potential solution to the long-standing problems of accurate and fast phase selection. This paper describes the design of a novel phase selector using ANNs. The technique is based on utilising fault generated high frequency noise (captured through the high voltage coupling capacitor of a conventional capacitor voltage transformer) to essentially recognise the various patterns generated within the frequency spectra of the fault generated noise signals on the three phases, for the purposes of accurately deducing the faulted phase. The paper demonstrates a new concept and methodology in phase selection which will facilitate single-pole autoreclosure applications in power systems.