With the large-scale commissioning of UHV AC and DC converter stations in China, the on-load tap changer (OLTC) has become one of the devices with more faults in UHV converter stations. To address the problem of difficult extraction of OLTC fault features in UHV converter stations in a high background noise environment, this paper proposes a method based on an adaptive particle swarm algorithm to optimize singular spectrum and singular value decomposition. First, the modal parameters in the singular spectrum decomposition algorithm are optimized using the APSO algorithm, and the optimal number of decomposition modes is selected. Secondly, the optimal singular spectrum components are selected based on the maximum cliff criterion. Then, the optimal reconstruction order is determined, and the signal is reconstructed by singular value decomposition, so as to achieve the purpose of signal noise reduction. Applying the proposed method to the simulated and experimental signals, the results show that the proposed method achieves a signal-to-noise ratio of 23.302, the root-mean-square error is only 0.004, and the waveform similarity parameter is as high as 0.998. This is better than other noise reduction methods. The method proposed can more effectively achieve the noise reduction of OLTC vibration signals in UHV converter stations, and provides a reference for auxiliary operation and maintenance personnel to diagnose the OLTC status.