The invention discloses a feature extracting method for power load dynamic features. The method comprises the steps of S1, preprocessing the three-phase transient state current data reflecting the power load dynamic features, and selecting the sampling data in one cycle before a fault and in two cycles after the fault as samples; S2, selecting a wavelet basis, and determining the lifting scheme of the wavelet basis; S3, selecting a resolving layer number m and performing m-layer resolving on the three-phase transient state current data of each sample by utilizing the wavelet packet conversion with the lifting scheme; S4, selecting the wavelet packet to reconstruct original signals, wherein the signal number in the wavelet packet is marked as r; and S5, extracting a corresponding reconstruction coefficient and calculating an energy value Mij(k) as the sample number, wherein j is equal to 1, 2 and 3 respectively representing the three phases of currents, k is equal to 1, 2, 3, ellipsis and r represents a frequency band. The multiplication operation number of the wavelet packet conversion based on the lifting scheme is reduced by about a half, so a faster conversion speed is realized.