Application of artificial neural network in noise mixed partial discharge recognition

Abstract

To test partial discharge (PD) recognition ability under different noise conditions, systemic research is carried out. In a noise-screened high voltage lab and using a high speed, wide-band digital measuring system, different kinds of PD current waveforms are recorded. Noises of different types are investigated. Then the PD signals are immersed into different noises with certain signal-noise ratios (SNR). By applying the segmented time domain data compression method, the features vectors of mixed waveforms are extracted. Employing a backpropagation algorithm, a feedforward triple-layered artificial neural network (ANN) program is generated and optimized. The mixed waveforms are tested and influence of each noise types in different SNR conditions are studied.