Abstract: InSAR technology is an important means and method to realize the analysis of surface subsidence in large-scale mining areas, and the accurate prediction of surface subsidence is of great significance for the prevention of geological disasters. Considering the strong volatility and nonlinearity of the surface subsidence data extracted from the mining area by InSAR technology and the difficulty of determining the hyperparameters of the long short-term memory (LSTM) network model, this paper proposed a LSTM prediction model for surface subsidence optimized by the variational mode decomposition (VMD) and the hunter-prey algorithm (HPO). Taking a specific mine as the research object, the subsidence information of the mining area was decomposed into multiple modal components using the VMD algorithm. Subsequently, the HPO-LSTM model was employed to predict these modal components, and the results were compared with the InSAR monitoring data. The results showed that the HPO-LSTM model exhibited the best prediction effect. The average absolute error was reduced by at least 85.41%, the root-mean-square error was reduced by at least 85.02%, and the average absolute percentage error was reduced by at least 87.05% compared with the VMD-BP and VMD-ELM models, which indicated that the proposed method was of more reliability and feasibility.