Abstract: Partial discharge signals are prone to missed detection under low signal-to-noise ratios, and the traditional singular value decomposition algorithm requires massive calculations when extracting partial discharge pulses. In this regard, a technology of partial discharge pulse extraction and denoising based on random singular value decomposition (RSVD) was proposed. It can extract partial discharge pulses and remove white noises. Compared with traditional SVD pulse extraction and calculation, the method requires a shorter time and has higher engineering practical value. Firstly, the sliding short-time data window was used to intercept the original partial discharge signal fragments. The maximum singular value calculated by RSVD was compared with the global optimal singular value threshold to determine the starting and ending points of pulses. Then the singular value decomposition (SVD) was combined with the local optimal singular value threshold to denoise the extracted signal. Experiments on typical simulated partial discharge pulses verified the superiority of the algorithm’s execution efficiency in pulse extraction. Partial discharge tests were performed on the cable defects simulated in the laboratory at power frequency voltage. Moreover, comparative experiments were conducted for the proposed method, discrete wavelet transform (DWT) plus the adaptive double-threshold method. The results show that the proposed method has a low missed detection rate and a remarkable performance in signal denoising.