Transformer fault diagnosis method based on two-dimensional cloud model under the condition of defective data

• Published in: Electrical engineering Vol. 106; no. 1; pp. 1 - 13
• Main Authors: Bai, Xingzhen, Zang, Yuan, Li, Jing, Song, Zhaoshan, Zhao, Kang
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.02.2024; Springer Nature B.V
• Subjects: Accuracy; Algorithms; Data processing; Economics and Management; Electrical Engineering; Electrical Machines and Networks; Energy Policy; Engineering; Fault diagnosis; Generative adversarial networks; Original Paper; Power Electronics; Redundancy; Self organizing maps; Transformers; Two dimensional models; Fault diagnosis; Transformer; Gas dissolved in oil; Defective data; Two-dimensional cloud model
• ISSN: 0948-7921; 1432-0487
• DOI: 10.1007/s00202-023-01964-7

In this study, facing the condition of structural and informational defects and complex correlation in data of fault feature gases for transformers, a transformer fault diagnosis (TFD) method based on the two-dimensional cloud model (2D-CM) under defective data is investigated. Firstly, to solve the problem that the accuracy of TFD is impacted by the completeness of the priori rule set caused by the incompleteness, imbalance and feature redundancy of priori transformer fault information, a data processing framework based on the strengthened Mahalanobis distance self-organizing map (SMSOM) and data generation algorithm is proposed. By applying the data generation algorithm based on the Wasserstein conditional generative adversarial network with gradient penalty and the filtering algorithm of the generated data based on SMSOM, the data set can be expanded evenly and the diversity of feature information in the limited training set can be maximized. Secondly, fault data is converted to cloud concept combination by constructing 2D-CM and adopting the computing method of cloud membership degree under offset space. Finally, TFD is realized by multi-rule fusion reasoning under the complete prior rule knowledge obtained by rule mining from all concept combinations. The experimental results indicate that the proposed data processing framework can make the TFD model based on 2D-CM achieve a diagnostic accuracy of 83.1%, which is superior to other data processing schemes. Moreover, the proposed TFD model based on 2D-CM increases the diagnosis accuracy by 22.8% compared with the traditional 1D-CM.