The present application discloses an OLTC (On-load tap-changer) signal denoising method based on STFT time-frequency spectrum coefficient shrinkage. The method comprises the following steps of: collecting vibration signals of an OLTC in a normal state by a vibration acceleration sensor; 2) performing STFT transformation of the collected OLTC vibration signals to obtain a time-frequency spectrum ofthe OLTC vibration signals; 3) employing a step size iterative algorithm to estimate an optimal threshold [Delta]*th in the interval from 0 to the maximum module value of a spectrum coefficient; and4) employing the optimal threshold [Delta]*th to perform spectrum coefficient shrinkage again to perform STFT inverse transformation for the obtained new time-frequency spectrum to reconstruct the signals after denoising. The OLTC signal denoising method can directly perform denoising for the impact feature signals mixed with noise so as to extract the impact components, and the mode is simple andintuitive, has strong pertinence, and can achieve better denoising effect.