Multiple classifier systems combined with localized generalization error for fault diagnosis of power transformers

Dissolved gas-in-oil analysis (DGA) is an effective approach for detecting incipient inner fault transformers and various methods derived from DGA have been introduced. To overcome their inherent weaknesses such as the variability of DGA data, this paper proposes a novel multiple classifier system to identify the inner fault of power transformers. The presented method is based on some primitive RBF classifiers and the multiple classifier system is evaluated with the Localized Generalization Error obtained by the Localized Generalization Error model (L-GEM). Compared to other measurements of ensemble system, the proposed method archives a good result.