Assisted Diagnosis of Real-Virtual Twin Space for Data Insufficiency

This paper proposes a virtual-real twin spatial fusion theory to deal with the problem of insufficient data for intelligent diagnosis of power equipment. First, build a transformer winding fault experiment platform, and use Frequency Response Analysis (FRA) to obtain the measured sample set that contains different fault locations, fault types and severity of the transformer to form an accurate physical space. Then, use COMSOL and matlab to build a transformer digital space model, correspondingly set the fault, obtain the simulation sample set, and form a fuzzy mirror space. Then, the sample data in the two spaces is pre-processed by feature extraction and fused with each other to obtain the virtual and real twin spaces. The fused samples are used as auxiliary training samples of intelligent fault diagnosis network. Finally, based on the proposed method, a variety of intelligent diagnostic networks are applied to diagnose the measured dataset, verifying that the proposed method can effectively improve the diagnosis and positioning effects of small sample data.