Abstract: This study conducts surface subsidence monitoring and predictive analysis for the subsidence caused by high-intensity mining in Renjiazhuang Mining Area. The small baseline subset-interferometric synthetic aperture radar (SBAS-InSAR) technique is adopted to invert the time-series deformation field of the mining area, realizing subsidence monitoring with millimeter-level accuracy. The sparrow search algorithm (SSA) is introduced to optimize hyperparameters and suppress noise interference. Combining the spatial feature extraction capability of the convolutional neural network (CNN) and the temporal pattern capturing capability of the long-short-term memory network (LSTM), a spatiotemporal collaborative prediction framework is constructed. The results indicate that: (1) During the monitoring period, the maximum cumulative subsidence in the mining area reaches-432.02 mm, and the maximum subsidence rate is -224.17 mm/a. (2) In the prediction of 6 typical monitoring points, the MAE and RMSE indicators of the fusion SSA-CNN-LSTM model are reduced by an average of 47.6% and 47.2% respectively compared with the LSTM model, and by an average of 34.3% and 37.2% respectively compared with the CNN-LSTM model, indicating a significant improvement in prediction accuracy.(3) The cumulative settlement at Point P6 in the special geological condition area is projected to reach-149 mm over the next 10 periods, posing significant safety hazards that require prioritized prevention and control measures. Through the "monitoring-prediction" system, this method improves the limitationsJP+1of traditional surface subsidence monitoring techniques, such as limited coverage and poor prediction capability. It provides technical support for harzard risk prevention and control in mining areas.