Research on power quality disturbance identification and classification technology in high noise background

• Published in: IET generation, transmission & distribution transmission & distribution, 2019-05, Vol.13 (9), p.1661-1671
• Main Authors: Li, Jianwen, Qin, Gang, Li, Yonggang, Ruan, Xiaofei
• Format: Article
• Language: English
• Subjects: characteristic curve extraction; complex matrix module; dynamic time warping distance query classification method; fluctuation energy density; matrix algebra; mean time‐frequency filter; multiresolution hyperbolic S‐transform; noise abatement; noise interference problem; noise suppression performances; nonsignal time‐frequency domain suppression; nontransient disturbance time‐frequency domain signal elimination; power quality disturbance classification technology; power quality disturbance signal identification; power supply quality; power system faults; power system transients; Research Article; signal time‐frequency domain database; time‐frequency analysis; time‐frequency filter denoising; transforms; transient disturbance time‐frequency domain; tree structure
• ISSN: 1751-8687; 1751-8695; 1751-8695
• DOI: 10.1049/iet-gtd.2018.6262

In order to solve the problem of noise interference in power quality disturbance identification, a new method of multiresolution hyperbolic S-transform for noise abatement based on energy density is proposed. First, multiresolution hyperbolic S-transform is performed on the power quality disturbance signal. This method combines suitable time-frequency resolution with good noise suppression performances. Second, the transient disturbance time-frequency domain is determined according to the fluctuation energy density. By using a mean time-frequency filter, the interference of the noise in the non-transient disturbance time-frequency domain with signals is eliminated, and the noise in the non-signal time-frequency domain is suppressed. Then, the signal time-frequency domain is determined according to the energy density, and the noise in the non-signal time-frequency domain is further suppressed by the denoising time-frequency filter. The characteristic curve is extracted from the complex matrix module after the noise abatement. Finally, a time-frequency database of tree structure is established. The dynamic time warping distance query classification method is used for quick classification according to the relationship of membership degree, which reduces the number of queries and improves the recognition accuracy. The classification result shows the effectiveness of the algorithm in high noise background and the applicability in actual fields.