Land subsidence is a common geological disaster, which seriously hinders the production and life of local residents. How to accurately predict land subsidence has become a hot topic discussed by relevant experts and scholars. But the conventional mathematical model is difficult to predict the land subsidence accurately. In this paper, the Sparrow Search Algorithm (SSA) is proposed to optimize the Elman land subsidence prediction method. At the same time, according to the principle of combination model, the strategy of SSA-Elman Residual Self-correction (SSA-Elman-RSC) model is proposed. The prediction error of neural network is reduced by residual correction, and the land subsidence prediction model is successfully applied to Xiaohe Industrial Park in Datong City, Shanxi Province. The prediction results are compared with those of the model without residual correction. The results show that SSA-Elman-RSC has higher accuracy for RMSE, MAE and MRE. The proposed model provides a new method for the prediction of land subsidence in Shanxi Province and a new idea for the establishment of combined model.