Pattern recognition of transformer partial discharge based on acoustic method

Online partial discharge pattern recognition can be applied to analyze the possible damages produced by partial discharge. Based on non-linear and non-stationary characteristics of partial discharge acoustic signal, fractal theory is adopted to extract fractal parameters from acoustic signal of transformer partial discharge in this paper. The fractal theory and the approach to calculating relevant parameters are introduced. Several typical kinds of partial discharge model are deigned in laboratory, which include point to plane discharge, void discharge, floating discharge and surface discharge. In order to simulate partial discharge in transformer, all these models are immersed into transformer oil. Acoustic signals of these models are obtained by digital measuring device. Then the fractal parameters (box counting dimension and lacunarity) are calculated, the calculated result show that fractal parameters are dissimilarity of different partial discharge acoustic signals. Finally these fractal parameters form the characteristic vectors, which are input into BP artificial neural networks for recognition. The result shows the fractal theory can be exploited to obtain the features of acoustic signals for accomplishing pattern recognition. Acoustic signal processing research provides us with a novel approach to study partial discharge, and it has a promising application future.