Partial discharge (PD) detection for power cable is an effective method in cable insulation diagnosis. Focusing on the deficiencies of white noise suppression method of traditional short-time singular value decomposition (STSVD), a modified de-noising method for partial discharge signal based on time-domain energy and adaptive singular value threshold is presented in this paper. The proposed method can estimate the number of reconstructed singular values precisely by using the adaptive singular value threshold estimation strategy, and the execution efficiency can be improved greatly combined with the time-domain energy criterion, which only handles the regions of partial discharge pulse. The proposed method has been examined on simulated and measured PD signals, and its results have been compared with existing adaptive singular value decomposition (ASVD), traditional STSVD, and wavelet transform de-noised results. The research results show that, compared with ASVD and wavelet transform de-noising methods, the proposed method can obtain better de-noised results; compared with traditional STSVD, the proposed method can effectively solve the pulse-like interference problem after de-noising, and the calculation rate is faster.