关键词: 矿区沉降, DS-InSAR, 预测模型, LSTM神经网络

Abstract: In response to the problem of low point density and uneven distribution in subsidence monitoring of mining areas using conventional InSAR technology, this paper uses 36 Sentinel-1A image data from August 2020 to August 2023 to obtain surface deformation information of Langyashan mining area in Chuzhou city, Anhui province using DS-InSAR technology. And the LSTM neural network model is used to predict the future settlement trend of the area with severe ground subsidence in the mining area, in order to understand the future development trend of ground subsidence in the mining area. The research results indicate that:①Compared with traditional PS-InSAR technology, DS-InSAR technology can significantly increase the number of monitoring points in mining areas and more comprehensively reflect surface subsidence information in mining areas. ②During the monitoring period, there are three deformation zones in the mining area, with a maximum settlement of 32.4 mm and a maximum settlement rate of 10.8 mm/a. ③By comparing with the GM (1,1) model and using the selected 6 settlement feature points, it is found that the LSTM neural network model exhibited higher prediction accuracy. ④For the area with the highest cumulative settlement, we use the LSTM model to predict the cumulative settlement of the 6 feature points in the area for the next 12 months. The prediction results show that the future settlement in the area fluctuates within a certain range, and no obvious settlement trend has been observed yet.

Key words: mining subsidence, DS-InSAR, prediction model, LSTM neural network