It is the key to realize reliable identification and diagnosis of discharge fault in distribution transformers in the study of acoustic characteristics of discharge fault noise and the noise of the distribution transformers itself and identification methods. We proposed a method for the diagnosis of discharge fault in distribution transformers based on complete ensemble empirical mode decomposition with adaptive noise（CEEMDAN）. Firstly

CEEMDAN was used to decompose the discharge fault acoustic signals

thus several intrinsic mode functions（IMF） could be obtained. The kurtosis of each IMF was sought to select the suitable IMF components for signal reconstruction and to extract the discharge acoustic signal. Then the CEEMDAN decomposition of the discharge acoustic signal was carried out again to obtain the marginal spectrum entropy

the center of gravity frequency

the band energy entropy and the singular spectrum entropy

which constitute the eigenvector. Finally

the support vector data description（SVDD） was used to classify and identify different typical discharge acoustic signals. The experimental results show that the proposed method can achieve more than 90% of the recognition rate when considering the superimposed noise of the distribution transformer

which can be used for the identification and diagnosis of the discharge fault of distribution transformers.