In response to the phenomena of significant oscillations and incomplete noise reduction when dealing with partial discharge signals using traditional methods, a combined approach based on complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and tunable Q-factor wavelet transform (TQWT) is adopted to denoise the PD signals. Firstly, the CEEMDAN is employed to decompose the noisy transformer PD signals into multiple intrinsic mode functions (IMF), and the correlation coefficient is utilized to assess the correlation between the IMF components and the original signal. Those with weak correlations are considered inferior IMFs. They are decomposed using TQWT. Energy proportion and kurtosis indicators are utilized to select wavelet sub-bands, extracting effective detailed information from the IMF. Subsequently, inverse transformation of the TQWT is applied to obtain new IMF components. The IMFs with strong correlations are considered high-quality. They are reconstructed together with the transformed new IMF components to obtain the denoising result. Simulation and field signal analysis verify the effectiveness and practicability of the proposed method. Compared to the traditional empirical mode decomposition(EMD) method, the percentage of waveform distortion decreased by 44.94% after denoising simulated signals using the proposed method. Compared to using only CEEMDAN, the noise suppression ratio increases by 26.64% after denoising on-site signals using the proposed method.