In order to extract the wind turbine bearing fault signal submerged under environmental and structural noise, a recursive variational mode decomposition method is proposed based on the energy tracking method, and the particle swarm optimization algorithm is used to solve the optimal constraint factor to obtain the accurate modal component. Based on the nonlinear fractal, the theory proposes a multifractal spectral feature factor to select the best modal component. Taking the fault degree and the loaded bearing acceleration signal as the object, the optimized recursive variational mode decomposition is used to obtain multiple components. The effective information component is selected by the maximum value of the multifractal spectral feature factor, and the fault classification is performed by the support vector machine. The results show that the optimized recursive variational mode decomposition can accurately decompose the vibration signal to different frequency bands for fault information extraction; the multifractal spectrum feature factor is positively correlated with the signal to noise ratio, and the component selected by its maximum value has more effective information; The BPNN is used to classified the hybrid fault degrees of different states, the test samples are constructed by selected components by IVMD-MFC with eight nonlinear characteristics. The diagnostic accuracy is 97. 5%. There is a good performance in hybrid fault degree of different status classification.