Abstract: In order to solve the problem that the vibration signal is easy to be interfered by noise and the state change is difficult to be perceived in real time when the transformer is running online, an online state perception method of transformer based on vibration signal decomposition is proposed. Firstly, the CEEMDAN is used to decompose the collected vibration signal to obtain the intrinsic modal components of different frequency scales. Secondly, the key features are extracted by wavelet packet energy analysis, and the state index is constructed to realize the unified quantitative expression of multi-dimensional features. Finally, the effectiveness of the proposed method is verified by experimental data. The results show that this method can effectively reflect the change of transformer operation state, improve the expression ability of vibration signal characteristics, and the average processing time of single window meets the requirements of on-line monitoring, which has good engineering application value.