InSAR技术结合深度学习的范围性地表沉降预测  

Prediction of range surface subsidence based on InSAR technology and deep learning

作　　者：杨帆[1] 胡晋 孙彩霞 YANG Fan;HU Jin;SUN Caixia(School of Geomatics,Liaoning Technical University,Fuxin,Liaoning 123000,China)

机构地区：[1]辽宁工程技术大学测绘与地理科学学院,辽宁阜新123000

出　　处：《测绘科学》2022年第7期60-68,134,共10页Science of Surveying and Mapping

基　　金：自然资源部国土卫星遥感应用重点实验室资助项目(LSMNR-202107);辽宁省教育厅项目(LJ2020JCL006);辽宁工程技术大学学科创新团队资助项目(LNTU20TD-27)

摘　　要：针对InSAR技术所提取的太原市局部地区沉降数据因其波动性和非线性影响,直接进行机器学习难以达到理想的预测效果的问题,该文提出一种自适应噪声完整经验模态分解(CEEMDAN)样本熵和深度学习组合的范围性地表沉降预测模型。基于CEEMDAN算法将沉降信息分解为多个模态分量(IMF)并计算其样本熵,利用相近原则对IMF重构后运用深度学习(Bi-LSTM)进行预测,将各序列的预测值叠加得到总沉降预测值。实验结果表明:相比BP神经网络、极限学习机(ELM)等模型,该预测模型的均方根误差、平均绝对误差和平均绝对百分比误差至少降低38.45%、41.26%和43.57%。表明该模型能够更好把握波动性较大的沉降信息,提高预测精度,为范围性的地表沉降预测提供了一种新的方法。Aiming at the problems of the fluctuation and nonlinear influence of the local subsidence data for Taiyuan City extracted by InSAR technology,it is difficult to achieve the ideal prediction effect by machine learning directly,this paper proposes an adaptive noise complete empirical mode decomposition(CEEMDAN)model based on sample entropy and depth learning.Based on the CEEMDAN algorithm,the model decomposes the settlement information into multiple modal components(IMF)and calculates the sample entropy.After IMF reconstruction,deep learning(Bi-LSTM)is used to predict the total settlement,and the predicted values of each series are superimposed to obtain the predicted value of the total settlement.The experimental results show that:compared with BP neural network,ELM et al,the root mean square error,average absolute percentage error and average absolute error of the prediction model are reduced by at least 38.45%,43.57%and 41.26%.It shows that the model can better grasp the fluctuating settlement information,improve the prediction accuracy,and provide a new method for the range of surface settlement prediction.

关 键 词：经验模态 样本熵 深度学习 地表沉降 INSAR 

分 类 号：P237[天文地球—摄影测量与遥感]