The broadband oscillation of a power system has the characteristics of wide frequency domain, nonlinearity and time variation. This entails higher requirements for the accuracy and rapidity of the oscillation classification. Therefore, a new broadband oscillation classification method based on adaptive variational mode decomposition (AVMD), multi-scale fuzzy entropy (MFE), and variable prediction model-based class discriminate (VPMCD) is proposed. First, AVMD is performed on the broadband oscillation signal to obtain the intrinsic mode functions (IMFS). Then, MFE is introduced to describe the time-domain features of the IMFS, and dimensionality reduction of the eigenvectors constructed by the IMFS is realized. Finally, the VPMCD is used to realize the classification detection of broadband oscillations on the feature vectors after MFE dimensionality reduction. The simulation and measured data analysis results show that the proposed method has higher detection accuracy and shorter classification time for broadband oscillation than that of the support vector machines (SVM) algorithm and the BP neural network algorithm.