Transformer Fault Diagnosis Algorithm Based on Entropy-Weighting Information Bottleneck Method

The paper presents an algorithm for DGA (Dissolved Gas Analysis) fault diagnosis based on the Information Bottleneck (IB) method by introducing the train data supervision after IB clustering. The classified label of the test data is marked by voting among all these train data which exist in the same cluster with the certain test data. Meanwhile, the paper proposes to apply the Entropy-weighting scheme to evaluate the important level of attributes. Experiment results show that the supervision IB method is feasible for the transformer fault diagnosis problem and the proposed algorithm is superior to Duval´s triangle method.