The rapid increase in the proportion of new energy sources causes great changes in the structure and stability characteristics of the system, and a new type of oscillation appears with multi-mode and broad-band characteristics. In order to predict, locate and suppress broad-band oscillation, it is necessary to identify the mode parameters of broad-band oscillation signals accurately and quickly. First, Both low noise and high noise signals are simultaneously decomposed by VMD, a criterion based on permutation entropy (PE) is used to distinguish between high and low noise signals. Then, improved variational mode decomposition (IVMD) is used to efficiently denoise high noise signals. Finally, hilbert transform (HT) is used to identify decomposed modes, and short-time fourier transform (STFT) provides time-frequency characteristics to help restore oscillations. The simulation and actual oscillation data verify that both obvious oscillation signals with low noise and ambient signals with high noise are effectively identified, and results reveal that the proposed method outperforms the other similar methods in accuracy and time of use.