The invention discloses a
partial discharge detection method based on ViT and Transform models, and the method comprises the following steps: obtaining a
partial discharge PRPD map, and converting the
partial discharge PRPD map into a feature PRPD image; using a pre-trained ViT model to extract
feature data of the feature PRPD image, and outputting the
feature data to a Transform model; a Transform model is self-defined, and
feature data are processed; and classifying the processed feature data by using global average
pooling and a full connection layer, and identifying various partial
discharge sources. According to the method, a pre-trained ViT model is used as a feature extractor, the partial
discharge PRPD atlas is converted into image data and feature data is extracted, a multi-layer structured TransformerBlock is arranged, the feature PRPD image can be better explained and analyzed, global average
pooling and a full connection layer further process the feature data, multi-
label classification is realized, and the accuracy of the partial
discharge detection is improved. The problems of pulse
signal overlapping and
noise signal overlapping of a plurality of partial discharge sources are solved, the method can adapt to PRPD image data of different sources and different formats, the detection speed and efficiency are improved, and the detection error is reduced.