The imbalanced class distribution among fault categories of power transformer leads to low diagnostic accuracy rate and the large difference in recognition effectiveness between fault types of traditional machine learning methods. Therefore

we propose a diagnosis model of multi-level hierarchical power transformer fault according to hierarchical classification and ensemble learning. In the model

classifiers are hierarchically constructed for level-by-level diagnosis in depth with consideration of the extent of imbalance at each level. The neural network classifier at the first level extracts three generalized feature tags

such as the normal

discharge fault

and overheat fault

to integrate with the original data input so as to guide classification among 9 exact operation statuses under DL/T 722—2014 standard. The classifier at the second level

EasyEnsemble learning

which generates balanced training subsets by undersampling the majority class

is adopted to synthesize the sub-classifier parameters for the ultimate classifier by parallel training

therefore

not only information between major fault types and minor ones is balanced

but also information loss in the undersampling process is prevented. Experiments show that the proposed model outperforms traditional methods with improvement in generalization ability of minor fault types and a 7% rise in diagnostic accuracy

giving a more accurate and balanced result on fault diagnosis of power transformers.