The invention belongs to the field of power systems, and relates to a PSO-based on improved variational mode decomposition. The LSSVM short-term load prediction method comprises the following steps: S1, selecting a decomposition effect evaluation index; S2, setting an SMD decomposition upper limit; S3, optimizing the VMD parameters by using a particle swarm optimization algorithm, performing VMD decomposition, and finally obtaining a period corresponding to the center frequency of the modal component; S4, combining the modal components to obtain a combined component; S5, solving mutual information between the sequences of the influence factor data and the combination components and the predicted daily load sequence, and obtaining an influence factor input variable set according to a threshold requirement; S6, substituting the selected influence factor input variable into the PSO- LSSVM model. According to the method, the utilization efficiency of influence factor data is improved, andan optimized mode decomposition result is obtained; By quantifying the correlation between the internal structure components of the influence factors and the loads, effective influence factor variables are accurately selected, the number of the influence factors is increased, and the prediction precision is improved.