In view of the non-stationary characteristics of rolling bearing vibration signals and the difficulty of extracting early fault features, a feature extraction method based on variational modal decomposition and mixing of time domain and frequency domain values was proposed, and the modified bat algorithm（MBA） was used to optimize support vector data description (SVDD) parameters for rolling bearing fault diagnosis. By using this method, the normal vibration signal was decomposed by variational mode decomposition and the modal function was obtained; singular value decomposition was used to extract the modal features of the modal function, and the time domain, frequency domain and modal features of the signal were extracted to construct the composite features (CF) to achieve 〖JP2〗the feature extraction; the modified bat algorithm (MBA) was used to optimize the kernel function width of SVDD, and the CF-MBA-〖JP〗SVDD fault diagnosis model was constructed. The model was used to diagnose various vibration signals of rolling bearings under different working conditions, and the overall fault identification rate was better than other comparison algorithms. The diagnosis and analysis of the whole life cycle bearing test data were conducted. The results show that by using this model, bearing faults can be early warned, which verifies the reliability and effectiveness of this method.