Monitoring and Prediction of Surface Subsidence in Mining Areas by Integrating SBAS-InSAR and ELM

With the rapid economic development in China, coal resources are being exploited greatly, which easily causes geological disasters due to surface subsidence. Fast and accurate surface subsidence monitoring and forecasting in mining regions are important references to analyze surface variation laws and disaster warning. However, differential interferometric synthetic aperture radar (D-InSAR) in mine surface monitoring is highly sensitive to spatiotemporal baseline and atmospheric delay. In addition, traditional machine learning algorithms have complicated network structures and difficulties determining parameters. Small baseline subsets InSAR (SBAS-InSAR) and extreme learning machine (ELM) dynamic prediction were combined for corresponding experimental studies to address these problems. On the basis of SBAS-InSAR, surface subsidence monitoring data in mining areas in Pingdingshan City, China, were collected, and a comparative analysis of D-InSAR monitoring data was performed, which verified the validity of SBAS-InSAR monitoring. On the basis of SBAS-InSAR data, a prediction model was built by ELM. The model results were compared with the prediction results of back propagation (BP) neural network and support vector machine (SVM) through root mean square error (RMSE) and mean relative error (MRE). Results demonstrate that the surface subsidence prediction of SBAS-InSAR in the monitoring mining area can reach millimeter accuracy. The MRE values of ELM, BP, and SVM prediction are maintained within 2%, 5%, and 8%, and the RMSE values are less than 3 mm, 7 mm, and 10 mm, respectively, thereby indicating that ELM prediction has high accuracy and reliability. This study provides an important evidence for safe production and scientific disaster prevention and reduction in mining areas.