Abstract: Aimed at the difficulties in extracting partial discharge fault information of transformers and low recognition rate of discharge types, a partial discharge pattern recognition method based on complementary ensemble empirical mode decomposition-multiscale permutation entropy (CEEMD-MPE) and stacked denoise auto-encoder (SDAE) is proposed. The CEEMD algorithm was used to decompose the original signals of partial discharge to obtain intrinsic mode functions (IMFs). According to the correlation coefficient, the IMF with the largest correlation coefficient was selected as the optimal component, and the permutation entropy (PE) value under different scales was calculated. The effective PE value was input as the feature dataset into SDAE for unsupervised learning. We use the Softmax classifier to output the discharge. The experimental results show that the algorithm recognition accuracy rate, recall rata and F1 score are 98%, 96.67% and 97.17% respectively, which can quickly and accurately recognize partial discharge types.