融合SBAS-InSAR技术与TSO-LSTM模型的矿区地表沉降预测方法  

Prediction Method of Surface Subsidence in Mining Area by the Integration of SBAS-InSAR Technique and TSO-LSTM Model

作　　者：肖海平[1] 夏益强 刘小生[1] 陈兰兰 XIAO Haiping;XIA Yiqiang;LIU Xiaosheng;CHEN Lanlan(School of Architectural and Surveying Engineering,Jiangxi University of Science and Technology,Ganzhou Jiangxi 341000,China;College of Resources and Architectural Engineering,Gannan University of Science and Technology,Ganzhou 341000,China)

机构地区：[1]江西理工大学土木与测绘工程学院,江西赣州341000 [2]赣南科技学院资源与建筑工程学院,江西赣州341000

出　　处：《金属矿山》2023年第1期126-133,共8页Metal Mine

基　　金：国家自然科学基金项目(编号:42171437);江西省自然科学基金项目(编号:20212BAB204030);江西理工大学高层次人才科研启动项目(编号:jxxjbs19032)。

摘　　要：矿区由于重工业器械的使用和采矿活动频繁,其岩层和地表容易发生沉陷和变形,快速、准确地分析、预测地表沉降是实现高效防灾减灾、推进绿色矿山建设的重要手段。针对现有预测模型监测点过少、多源数据难以获取以及网络模型超参数难以确定等问题,提出了一种基于金枪鱼群(Tuna Swarm Optimization,TSO)优化长短时间记忆(Long Short-Term Memory,LSTM)网络模型超参数的深度学习预测方法,利用多个高相干性点的沉降时序实现矿区的精准预测。利用SBAS-InSAR技术处理50景覆盖德兴铜矿区的Sentinel-1 A升轨SAR影像,获取了该区域25465个高相干性点的沉降时间序列。利用TSO算法优化LSTM网络模型超参数,寻找出最适合该矿区沉降时序预测的LSTM网络模型,并使用优化后的LSTM网络模型分区域对沉降区开展沉降时序预测并计算预测精度。研究表明:使用TSO算法优化LSTM网络模型超参数是有效的,优化后的模型均方根误差至少降低了20%,平均绝对值误差至少降低了35%,预测均方根误差不超过2 mm,预测平均绝对误差不超过3 mm,模型平均预测精度超过95%。所提方法为确保安矿区全安全生产,实现科学防灾、减灾提供了技术支持。Due to the use of heavy industry equipment and frequent mining activities in mining areas,the rock strata and surface are prone to subsidence and deformation.Analyzing and predicting the surface subsidence quickly and accurately is the key to efficiently prevent and reduce disaster and promote the construction of green mines.In view of the problems,such as too few monitoring points of existing prediction models,difficult to obtain multi-source data and difficult to determine the hyperparameters of network models,a deep learning prediction method based on Tuna Swarm Optimization(TSO)is proposed to optimize the hyperparameters of Long Short-Term Memory(LSTM)network model,which uses the subsidence time series of multiple high coherence points to achieve accurate prediction of mining areas.In this paper,SBAS-InSAR technique is used to process 50 scenes of Sentinel-1A ascending SAR images covering Dexing Copper Mining Area,and the subsidence time series of 25465 high coherence points in this area is obtained.The TSO algorithm is adopted to optimize the hyperparameters of the LSTM network model to find the most suitable LSTM network model for the subsidence time series prediction of the mining area.The optimized LSTM network model is used to predict the subsidence time series in the subsidence area and calculate the prediction accuracy.The study results show that it is effective to use TSO algorithm to optimize the hyperparameters of LSTM network model.The root mean square error(RMSE)of the optimized model is reduced by at least 20%,the average absolute value error(MAE)is reduced by at least 35%,the root mean square error of prediction is less than 2 mm,the average absolute error of prediction is less than 3 mm,and the average prediction accuracy of the model is more than 95%.The proposed method provides technical support for ensuring safe production and realizing scientific disaster prevention and reduction in mining area.

关 键 词：开采沉陷 深度学习 金枪鱼群优化 长短时间记忆 沉降预测 SBAS-InSAR TSO-LSTM 

分 类 号：TD325[矿业工程—矿井建设] P237[天文地球—摄影测量与遥感]