New features derived from dissolved gas analysis for fault diagnosis of power transformers

Dissolved gas analysis (DGA) is usually used for fault diagnosis of filed transformers. However, the gas contents dissolved in oil are easily influenced by transformer structure and capacity, fault location and fault severity, etc., which would reduce the reliability of transformer fault diagnosis. In order to improve the accuracy of transformer fault diagnosis, a new group of DGA-derived features was proposed based on the selection by support vector machine (SVM) and genetic algorithm (GA). First of all, using 28 DGA ratios as the input vectors, the fault diagnosis model of transformers was designed based on SVM. Then, the SVM parameters and DGA ratios were simultaneously optimized by GA achieving the 9 selected DGA ratios as the new features for transformer fault diagnosis. The diagnosis results on the data from IEC TC 10 database show that the proposed DGA ratios achieve the fault diagnosis accuracy of 84%, which increase the accuracy by 10%-25% over the frequently-used DGA data and IEC ratios. In addition, no matter what kind of features are adopted, the diagnosis results of SVM are superior to that of the frequently-used neural network. Finally, the diagnosis results of domestic 117 transformers using the new DGA ratios are determined as 87.18% which verify the effectiveness of the proposed method once again.