The noise mixed in the voltage sag signal will obscure the start and end time of the voltage sag in detec-tion

and the interference of the voltage fluctuation will affect the determination of the of the voltage sag depth. Thispaper proposes a denoising method based on the variational mode decomposition（VMD）

and designs a TDM（thirdderivative method）-Prony algorithm combined with wavelet transform. First

the signal is decomposed via VMD to ob-tain several band-limited intrinsic modal components（BLIMFs）

and the acquired high-frequency BLIMFs are de-noised by SURE threshold then reconstructed with the low-frequency components. Then

wavelet transform is per-formed on the reconstructed signal to locate the start and end time of voltage fluctuation and voltage sag

and TDM-prony algorithm is used to fit the fluctuation signal segment to obtain its information

so as to achieve the separation offluctuation and sag. Finally

Hilbert transform is performed on the clean sag signal to obtain the amplitude and dura-tion of voltage sag. Simulation results show that the designed algorithm can effectively extract the feature of voltagesag interfered by fluctuating signals.