Discrimination of phase-shifted partial discharge patterns by neural network using standardization method

In this work, a neural network algorithm using the standardization method was developed to discriminate phase-shifted partial discharge (PD) patterns. Previous efforts to recognize PD patterns have been performed by applying the neural network to the φ-q-n pattern, which is composed of phase angle φ, discharge magnitude q and repetition rate n of PD pulse. Considering PD measurement in the field, it is not so easy to acquire the information on absolute phase angles of PD pulses. As a consequence, one of the significant aspects in applying the neural network algorithm to practical systems, is to develop a method that can discriminate the phase-shifted φ-q-n patterns. Therefore, the authors have established a new method which can convert the phase-shifted φ-q-n patterns to a standardized φ-q-n pattern. This standardized φ-q-n pattern could be obtained by applying Fourier analysis and was not influenced by the phase-shift of PD patterns. As a result, the recognition performance of the neural network for the phase-shifted φ-q-n patterns was considerably improved by applying this new standardization method