摘要
针对传统矿区开采沉陷监测方法耗费人力财力和预测预警模型较少的问题,研究提出一种基于短基线集合成孔径雷达干涉测量(Small Baseline Subset-Interferometry Synthetic Aperture Radar,SBAS-InSAR)技术和卷积神经网络(Convolutional Neural Networks,CNN)与长短期记忆网络(Long Short-Term Memory,LSTM)相结合的矿区开采沉陷监测预测方法。首先,利用SBAS-InSAR技术对建新煤矿进行矿区开采沉陷监测,获取了该矿区的年平均沉降速率和累计沉降值。用GNSS监测数据与SBAS-InSAR结果进行对比验证,其拟合效果较好。其次,在此基础上利用CNN-LSTM模型预测后6期沉降数据,其结果与CNN和LSTM预测结果进行对比。研究显示,CNN-LSTM模型的平均绝对误差(S_(MAE))和均方根误差(S_(RMSE))比单一的CNN和LSTM分别至少降低了44.8%和40.6%,其决定系数均高于98%。最后,进一步预测前6期和中6期沉降数据,验证了CNN-LSTM预测模型在时间上的一致性。因此,SBAS-InSAR融合CNN-LSTM模型在类似矿山开采沉陷监测和预测中有较好的应用前景。
In response to the limitations of conventional mining subsidence monitoring methods,which often require significant human and financial resources and lack robust prediction and warning models,this paper presents an innovative approach.It combines Synthetic Aperture Radar Interferometry(SBAS-InSAR)technology with Convolutional Neural Network(CNN)and Long Short-Term Memory(LSTM)networks to develop a predictive framework for monitoring mining subsidence.To begin,40 observations of Sentinel-1A image data captured between September 4,2021,and September 18,2023,were analyzed using SBAS-InSAR technology to monitor the subsidence of Jianxin Coal Mine.This analysis yielded the annual average subsidence rate and cumulative subsidence value for the mining area.The findings indicate a maximum annual average subsidence rate of 131 mm/a and a maximum cumulative subsidence of 379 mm over the study period.A comparison between GNSS monitoring data and SBAS-InSAR results demonstrates a satisfactory correlation.To address the absence of some images in the study area,a three-time spline interpolation method is employed.This yields a comprehensive time series dataset,comprising 63 intervals,capturing the cumulative morphology of the study area at 12-day intervals from September 4,2021,to September 18,2023.Initially,the first 57 periods of data are allocated for training,while the remaining six periods are designated for testing.Subsequently,the CNN-LSTM model is deployed to forecast the sedimentation data for these six periods,and the outcomes are contrasted with the predictions from the individual CNN and LSTM models.Notably,the CNN-LSTM model exhibits a minimum reduction of 44.8%in Mean Absolute Error(MAE)and a decrease of 40.6%in Root Mean Square Error(RMSE)compared to the standalone CNN and LSTM models,respectively.Additionally,both the coefficients of determination surpass 98%.Subsequent predictions of the first six and middle six periods of subsidence data affirmed the temporal consistency of the CNN-LSTM prediction model.Conse
作者
师芸
折夏雨
张雨欣
王凯
张琨
吴睿
SHI Yun;SHE Xiayu;ZHANG Yuxin;WANG Kai;ZHANG Kun;WU Rui(College of Mapping Science and Technology,Xi'an University of Science and Technology,Xi'an 710054,China;Key Laboratory of Coal Resources Exploration and Comprehensive Utilization,Ministry of Natural Resources,Xi'an 710021,China;College of Architecture and Civil Engineering,Xi'an University of Science and Technology,Xi'an 710054,China;Guangzhou Southern Surveying and Mapping Technology Co.,Ltd.,Guangzhou 510000,China)
出处
《安全与环境学报》
CAS
CSCD
北大核心
2024年第9期3429-3438,共10页
Journal of Safety and Environment
基金
国家自然科学基金项目(42174045)。
