In view of high noise interference and the difficulty of useful signal extracted for direct-driven permanent magnet motor in coal mine, a denoising method of integrating improved VMD and wavelet soft threshold is proposed. Firstly, particle swarm optimization is used to optimize the decomposition layers K and penalty factor α of the variational modal decomposition algorithm to obtain the optimal parameter combination. Based on the optimal parameter combination, K eigenmode components of abnormal sound signal for directdriven permanent magnet motor in coal mine are obtained. Secondly, the weighted margin index is used to screen out the effective signal components and the noisy components that need further decomposition. The wavelet soft threshold is used to further denoise the noisy components. Finally, the effective signal component and the wavelet soft threshold denoised component are reconstructed to obtain the final denoised signal. This method is used to denoise the simulation signal and the abnormal sound signal of direct-driven permanent magnet motor in coal mine respectively. In order to prove the validity of the method, we conduct the comparative test. The test results show that this method can increase the SNR of simulation signals to 27. 524 7 dB, and reduce the root mean square error to 0. 085 5. The SNR of measured signals is improved to 34. 715 3 dB, and the root mean square error is reduced to 0. 006 7. The method proposed can denoise effectively and provide data basis for subsequent fault feature extraction and fault diagnosis.