Spatiotemporal Evolution of Surface Subsidence in Large-Scale Mining Areas Under Rainfall Influence and Optimization Model Development

Considering the challenges of traditional monitoring methods in achieving large-scale surface subsidence monitoring over mining areas, as well as the difficulties in modeling settlement prediction area, methods and acquiring model hyperparameters, this paper integrates rainfall data from the mining analyzes the spatiotemporal evolution characteristics of surface subsidence using SBAS-InSAR technology, and proposes an APO-BiLSTM settlement prediction model. This model employs the Arctic Puffin Optimization (APO) to optimize the hyperparameters of a bidirectional long short term memory (BiLSTM) network. The research results indicate that rainfall has caused the formation of nine distinct subsidence areas in the mining area, with Subsidence Area IX experiencing the most severe subsidence, covering an area of 9.31 km², with an average annual subsidence rate as high as -331 mm/a and a maximum cumulative subsidence of 427 mm. In the early stages of subsidence, a “subsidence-lifting-subsidence-lifting” phenomenon is observed, which gradually stabilizes in the later stages. Additionally, compared to the LSTM and BiLSTM models, the proposed APO-BiLSTM model reduces the root mean square error (RMSE) of single-step predictions by 79.8% and 76.6%, respectively, and the mean absolute error (MAE) by 79.1% and 75.9%, while increasing the R² by 6.0% and 4.4%. The absolute error of 78.3% of the high coherence points is less than 4 mm, indicating that the model has promising application prospects in large-scale surface subsidence prediction in mining areas.