The discharge sound signal has been widely utilized to detect the partial discharge of the ring network cabinet, which has the characteristics of rich information and can accurately reflect the discharge fault. However, the effective detection of the discharge sound signal is a difficult point. The two-channel recording has unique advantages in the field of noise cancellation, which can effectively eliminate non-stationary noise. According to the characteristics of long-distance non-stationary noise measured on-site, the main noise interference has long-distance non-stationary feature. But, the partial discharge signal is a short-distance sound source. The noise elimination method of the discharge signal of the ring main unit is based on the dual-channel energy difference (power level difference, PLD). Compared with the single-channel denoising method based on spectral subtraction and Wiener filtering, the simulation experiment results show that the signal-to-noise ratio of the proposed denoising method under the two non-stationary noises is 14. 8 dB and 9. 1 dB higher on average, respectively. The square errors ( 1 × 10 -4 ) is reduced by 19. 34 and 15. 50 on average, respectively, and the denoising performance is better than that of the single-channel denoising algorithm. In the field experiment, the discharge sound waveform of the ring network cabinet is enhanced, which can effectively remove non-stationary noise of the surrounding environment. Thus, the effective discharge signal is retained, which can provide effective data support for the partial discharge diagnosis of the ring main unit.