Abstract: In a complex noise environment, the high completeness extraction of partial discharge (PD) signals of power equipment is the key to the online evaluation of its status. This paper proposes a method for extracting complex noisy PD signals based on adaptive complete ensemble empirical mode decomposition (CEEMDAN) and wavelet packet. Firstly, the noisy PD signal is decomposed by adaptive CEEMDAN, and the narrowband noise and frequency aliasing contained in the component are suppressed by using singular value decomposition (SVD) algorithm. Then, the effective components are determined according to the correlation coefficient to reconstruct the signals. Finally, the modified wavelet packet threshold method is employed to filter the white noise in the reconstructed signal. The algorithm is used to denoise the simulated data and the measured data separately. The quantitative analysis results show that the method can effectively remove white noise and narrowband noise interference. The waveform of the extracted PD signal has small distortion and energy loss, which can meet the subsequent application.