Hybrid method on signal de-noising and representation for online partial discharge monitoring of power transformers at substations

To ensure reliable operations of power transformers, online condition monitoring needs to be performed. However, extensive noise can be coupled into measured signals and cause ambiguities in evaluating transformers' conditions. This study proposes a hybrid method, which combines pre-whitening and blind equalisation for de-noising the signals obtained from online partial discharge (PD) measurements of transformers. A measured signal is first gone through a pre-whitening process for initial noise reduction and then processed by blind equalisation. Finally, an equalised signal that can reveal PD source in a transformer is converted to a kurtogram for an accurate PD pattern representation. The proposed method has been applied to signals obtained from laboratory experiments and online measurements of transformers at substations. Results show that the method can effectively de-noise PD signals contaminated by severe noise and consistently represent PD patterns induced by different PD sources.