Partial Discharge Denoising And Pattern Recognition Based On Singular Value Decomposition

Partial discharge(PD)is a manifestation of the insulation aging of electrical equipment.Therefore,PD inspection can be used to evaluate the insulation status of equipment and find internal defects in electrical equipment.The study of PD signal denoising and pattern recognition is beneficial to PD signal detection and analysis.Based on singular value decomposition(SVD),this paper uses the differences of singular values of different signals,the good robustness of singular values in reflecting the inherent characteristics of matrix,and the insensitivity of singular values to noise disturbance transformation to complete the denoising and pattern recognition of PD signals.Aiming at the narrow-band interference suppression,this paper proposes a narrow-band interference suppression method based on Hankel matrix and SVD.Firstly,Hankel matrix is selected as the trajectory matrix for SVD,and the results show that the singular value of narrow-band interference is twice of the number of frequencies,and the singular value of narrow-band interference is greater than that of PD and white noise.Then,the effective order of the singular value of the narrow-band interference is determined by the singular value difference spectrum method to reconstruct narrow-band interference,and then the reconstructed signal is subtracted from the original signal to suppress narrow-band interference.The effectiveness of the method is verified by denoising the simulated signal and the measured signal.Aiming at the white noise suppression,this paper proposes a local discharge white noise suppression method based on S-transform and SVD according to the characteristics that white noise obeys?~2 distribution in S-transform domain and its average power spectrum is proportional to frequency.Firstly,the time-frequency matrix of S-transform is taken as the track matrix of SVD,and the time and number of PD are determined by SVD,then the time-frequency matrix is processed by hard threshold method to complete the first denoising,and the second denoising is completed by SVD for the residual white noise.The effectiveness of the method is verified by denoising the simulated signal and the measured signal.For PD pattern recognition,this paper proposes a PD signal recognition method based on SVD and probabilistic neural network.Firstly,This method analyzes that the effective information of PD signal can be reflected by less singular values in the time when PD occurs.The percentage of singular value energy is selected as the eigenvector,and the pattern recognition of four different types of PD signals in cables is well completed by GA-PNN.