Hierarchical Diagnosis Method for Transformer Faults Driven by Mixed Data and Experience

The early warning and diagnosis of transformer fault are important to the safe and stable operation of a power system. Based on multi-dimensional data including online monitoring, operation, maintenance, and testing, this paper proposes a hierarchical diagnosis method for oil-immersed transformer faults driven by mixed data and experience. Firstly, based on the online and offline data of dissolved gas in oil, a primary fault diagnosis model based on deep forest is constructed for the predictions of six fault properties. Then, combined with transformer fault trees and relevant criteria, the classification of fault properties and fault types is realized. The association rules are introduced to analyze the correlation between fault types and features, and several features are extracted to eliminate specific fault types. A refined transformer fault diagnosis model covering 21 fault types is built, and the dynamic evaluation of fault types, location, and credibility is realized. Finally, based on single and group samples composed of 500 kV transformers, the validation of the fault diagnosis method is carried out. The diagnosis results of individual transformer show that the method can be employed to effectively identify the fault types and location. The accuracy of diagnosis results based on group samples is 89.0% and it is also applicable to the diagnose multi-fault transformers. The method realizes the concretization and quantification of transformer fault diagnosis results, and it can provide targeted guidance for transformer operation and maintenance. © 2023 Science Press. All rights reserved.