Research Of Transformer Differential Protection Based On Wavelet Transform

In the context of rapid economic development, the requirement of electricity has been unceasingly increasing, the scale of power system is expending continually. As the important equipment in power system, the function of transformer is more and more prominent, higher reliability of transformer is required by customers. Because of the simplicity and high efficiency, the differential protection is widely used in transformer protection, but when the magnetizing inrush current is produced, the differential protection will maloperation. Therefore, the accurate identification of the magnetizing inrush current is the research focus to ensure reliable operation of the transformer.The wavelet transform, which has been developed rapidly in recent years, is a mathematical tool developed by the Fourier transform. Wavelet transform overcomes the deficiencies of the Fourier transform, it is good at the time-frequency domain localization, and can extract the characteristic of the singular signal accurately. Compared to the internal fault current, the magnetizing inrush current has obvious singularity, it can be extracted by wavelet transform accurately. According to the generation mechanism of the magnetizing inrush current and internal fault current, a new method-wavelet energy moment method, based on the ratio of the change rate of harmonic, is described to identify the magnetizing inrush current.In the paper, Matlab is used to establish the simulation models, and the simulation data is processed by the’db2’wavelet. The wavelet energy moments of high harmonics within each scale are analyzed, to get the change rate of energy moment, which means the changes of harmonic content in each band. We use the ratios of the change rate, labeled by ’K’, to distinguish the magnetizing inrush current and internal fault current correctly. The threshold of K which labeled by ’M’ is got by large number simulations. The simulations shows that the ratio K of inrush current is lower than M, the ratio K of internal fault current is larger than M, this method needs only one cycle of sampling data to make the correct identification.