Abstract: The information flow of virtual power plants (VPPs) can accurately represent the level of interaction between distributed energy resources (DERs) and the power system, and precise prediction of this information flow is crucial for understanding the operational characteristics of VPPs and improving the efficiency of system control. However, VPPs involve multiple interacting entities, and their energy characteristics are influenced by both the stochasticity of individual resources and the aggregate characteristics of a large number of resources. Therefore, traditional prediction models struggle to effectively fit the information flow of VPPs, necessitating the development of a model specifically tailored for predicting VPP information flow. This paper proposes a short-term hybrid prediction model based on long short-term memory (LSTM) and variational mode decomposition (VMD). The model first utilizes VMD to extract intrinsic mode components of the flow sequence based on the variational principle. Then, LSTM is employed to model and predict each mode component sequence separately, and an attention mechanism is innovatively introduced to select important feature sequences from the intrinsic mode components. Finally, all sub-prediction algorithms are integrated into a complete prediction model. Simulation results demonstrate that compared to traditional methods, this model can effectively improve the accuracy of VPP information flow prediction.