【摘要】 为研究地下采矿引发的地面沉降预测问题，以山西省晋城市阳城县为背景开展地表沉降监测与预测方法研究。首先，获取2018年1月至2020年12月期间的Sentinel-1 SAR影像(81景),结合数字高程模型(DEM)、大气校正在线服务(GACOS)及精密轨道数据，采用小基线集干涉合成孔径雷达(SBAS-InSAR)技术精细化监测区域地表形变情况，揭示其时序演化与空间分布特征(最大沉降速率达27.84 mm/a);然后，构建基于变分模态分解(VMD)与反向传播(BP)神经网络相结合的混合预测模型(VMD-BP);最后，将该模型预测性能与传统长短期记忆网络(LSTM)模型及变分模态分解与长短期记忆网络(VMD-LSTM)模型进行对比分析。结果表明：VMD-BP模型显著提升了预测精度，在测试点位(点位a)的均方根误差(RMSE)、平均绝对误差(MAE)和平均绝对百分比误差(MAPE)分别低至0.278 01 mm、0.234 29 mm和0.39%,远优于LSTM及VMD-LSTM模型。更多还原

【Abstract】 In order to study the problem of surface subsidence caused by underground mining, a study on the methods of surface subsidence monitoring and prediction was carried out in Yangcheng County, Jincheng City, Shanxi Province. Firstly, Sentinel-1 SAR images(81 views) from January 2018 to December 2020 were acquired, combined with digital elevation model(DEM), generic atmospheric correction online service(GACOS), and precision orbit data, and SBAS-InSAR technique was used to monitor the regional surface deformation in a refined way, revealing its temporal evolution and spatial distribution characteristics(the maximum subsidence rate reached-27.84 mm/a). Then, a hybrid prediction model(VMD-BP) based on the combination of variational modal decomposition(VMD) and back propagation(BP) neural network was constructed; finally, the prediction performance of this model was compared with that of the traditional LSTM model and VMD-LSTM model. The results show that the VMD-BP model significantly improves the prediction accuracy, and the root mean square error(RMSE), mean absolute error(MAE), and mean absolute percentage error(MAPE) at the test point(point a) are as low as 0.278 01 mm, 0.234 29 mm, and 0.39%, respectively, which are much better than the LSTM and the VMD-LSTM models.更多还原