A Top-down Approach to Partial Discharge Recognition of Current Transformers

This paper presents a partial discharge (PD) recognition procedure. The top-down approach discussed first how to make the tailor-made current transformer (CT) models, how to obtain the 3D patterns from the CT models in a magnetically shielded room, and finally how to apply of an artificial neural network (BPN) in recognition. Firstly, tailor-made cast-resin current transformers with insulating defects were made for testing. The testing used magnetically shielded room and a commercial PD detector system to obtain 3D patterns of four experimental models for recognition niche. Secondly, through preprocessed data originating from the detecting system, training data sets for a back-propagation artificial neural network are used to be PD recognition patterns in three kinds of defects of current transformers and in a perfect one. Finally, with a view to exploring applicability in the field, this research randomly selects different levels of noise to distort the original training and testing set. These distorted data sets are entered for testing. Research results show that, with 20% noise per discharge count, an 80% successful recognition rate is achieved.