When the fan is running under various working conditions, the signal detected by the vibration monitoring system of the bearing is difficult to realize fault diagnosis, and the fault diagnoses of the bearing studied in a large number of literatures are mostly carried out at constant speed. Aiming at the rolling bearing running under variable working conditions, a feature extraction method based on SHO-VMD decomposition and multi-feature parameter fusion was proposed. The t-SNE dimension reduction visualization was used to extract the fault information of the vibration signal and the speed change information. The decomposition effect of variational mode decomposition (VMD) method depends on the number of decompositions and the value of penalty factors. The parameters were optimized by the selfish sheep swarm optimization (SHO) algorithm. The vibration signal was decomposed into some intrinsic mode components, and then the characteristic parameters of each component were extracted. The multi-feature fusion was carried out based on singular value feature, energy entropy and sample entropy. The t-SNE dimension reduction was used to extract the fault information and speed change information of the bearing. The experimental results show that the proposed method can be used to effectively extract the fault and speed information of the bearing.