Abstract: To solve the problems existing in the swarm optimization algorithm, such as the low efficiency in searching the modal number and quadratic penalty parameters for the variational mode decomposition, a signal pretreatment method was proposed for fast local mean empirical mode decomposition. The number of modal was estimated according to the similarity coefficient criterion, and the penalty parameter was selected with the multi-evaluation factors. In order to reduce the influences of high natural frequencies band on results of VMD, the signal was reconstructed based on autocorrelation energy function to reduce the influence of noise and high frequency band. Simulation test and real data analysis of bearing fault were carried out to compare the parameter selection effect with swarm optimization algorithm and VMD modal selection effect based on similar center frequency. The results show that this method can extract early weak fault characteristic frequencies of bearings effectively.