Effective noise reduction of power quality(PQ)signals is the basis of detection and recognition of PQ signals. The block-matching and 3D collaborative filtering(BM3D)algorithm is at present the most efficient de-noising algorithm for Gauss noise and other noise models in the field of image processing. To overcome the difficulty of one-dimensional PQ signals de-noising

that is

effectively removing noise and well keeping singularities intact

the BM3D algorithm is improved and a novel adaptive noise reduction method for PQ disturbances is proposed. For the proposed method

as less parameters are used

there is no need to estimate noise variance

nor is it necessary to artificially set the filter thresholds. Rather

accurate thresholds are calculated adaptively for the discrete cosine transform(DCT)domain filtering to obtain the de-noising results of PQ disturbances. The simulation experiments of six common kinds of PQ disturbances

including voltage interruption

voltage sag

voltage swell

impulsive transient

oscillation transient and harmonic

are performed and a comparison with the widely used wavelet threshold de-noising method is analyzed. Finally

the actual PQ data is employed. The experimental results indicate that the proposed method is effective. This work is supported by Fundamental Research Funds for the Central Universities(No. 2682014RC07).