Due to the lack of typical damage samples in the transformer fault diagnosis

a new method based on chaos support vector machines（CSVMs） was proposed.According to the method

the five characteristic gases dissolved in transformer oil were extracted by the K-means clustering（KMC） method as feature vectors

which were input to chaotic optimal multi-classified SVMs for training.Then the CSVMs diagnosis model was established to implement fault samples classification.The experiment shows that by adopting facture extraction with KMC

the diagnosis information is concentrated and the time-consuming in parameter determination is solved effectively.On the other hand

chaos optimization better enhanced model extension ability.Moreover

the presented method enables to detect transformer faults with a higher correct judgment rate

and can be used as an automation approach for diagnosis under condition of small samples.