Partial discharge de-noising employing adaptive singular value decomposition

Noise reduction is one of the important tasks of partial discharge (PD) detection and investigations in the process of cable insulation assessment. Singular Value Decomposition (SVD) is an algorithm, which can be employed in noise reduction from partial discharge signals, based on separation of singular values corresponding to noise and PD signal. In this paper, a new idea is presented providing generalization and practicality of the method in selection of singular values for detection and/or reconstruction of de-noised PD pulses. This procedure named as Adaptive Singular Value Decomposition (ASVD) has been examined on field measured, laboratory recorded and simulated signals and its results have been compared with the results of some published methods. The proposed method cleans up polluted PD pulses, even in excessive noisy conditions, resulting in higher signal-to-noise ratios (SNRs) rather than the results of other mentioned methods. Comparing in terms of magnitude, falltime, and pulse-area of PD signals after de-noising, the proposed method also preserves the features of PD pulses from distortions and errors much better than the compared methods.