小波-LSTM神经网络在地铁沿线沉降预测中的应用  

Application of wavelet-LSTM neural networks for subsidence prediction along metro lines

作　　者：孙宇超 魏长寿 李志进 张明刚 刘玉针 SUN Yuchao;WEI Changshou;LI Zhijin;ZHANG Minggang;LIU Yuzhen(College of Geodesy and Geomatics,Shandong University of Science and Technology, Qingdao Shandong 266590, China;School of Mining and Coal, Inner Mongolia University of Science and Technology, Baotou Inner Mongolia 014010,China)

机构地区：[1]山东科技大学测绘与空间信息学院,山东青岛266590 [2]内蒙古科技大学矿业与煤炭学院,内蒙古包头014010

出　　处：《北京测绘》2022年第5期664-669,共6页Beijing Surveying and Mapping

摘　　要：为缓解城市交通压力,地铁工程的修建日益加快,但其施工、运营都会造成沿线地表沉降,为有效预防地表沉降引起的地质灾害。本文基于51景升轨Sentinel-1A卫星影像,应用差分干涉测量短基线集时序分析(SBAS-InSAR)技术获取青岛地铁三号线沿线地表形变信息,分析地铁沿线主要沉降区域的成因,并对沉降区域内的特征点使用小波分解、重构,对降噪后的形变时间序列进行了模拟和预测。发现了4个主要的沉降区域,其中青岛北站周边沉降最为严重,沉降速率为-10.42 mm/a。优化后的长短期记忆(LSTM)神经网络模型对形变时间进行预测,其精度比传统LSTM、多层前馈BP神经网络模型更优,证明该模型在城市地铁沿线的地质灾害预防中具有广泛应用价值。In order to relieve the pressure of urban traffic,the construction of metro projects is accelerating,but its construction and operation will cause ground settlement along the line,in order to effectively prevent the geological disasters caused by ground settlement.This paper applies small baseline subset-interferometric synthetic aperture radar(SBAS-InSAR)technology to obtain surface deformation information along Qingdao Metro Line 3 based on 51-view uplift Sentinel-1A satellite images,analyses the causes of major subsidence areas along the metro line,and simulates and predicts the deformation time series after noise reduction by using wavelet decomposition and reconstruction for the feature points within the subsidence areas.The optimized long short-term memory(LSTM)neural network model predicted the deformation time series with better accuracy than the traditional LSTM and back propagation(BP)neural network models,which proved that the model is widely used in the prevention of geohazards along urban metro lines.

关 键 词：差分干涉测量短基线集时序分析(SBAS-InSAR)技术 小波降噪 长短期记忆(LSTM)神经网络模型 形变预测 

分 类 号：P237[天文地球—摄影测量与遥感]