门控循环神经网络的时序PS-InSAR地面沉降预测  

The prediction of the time-series PS-InSAR ground subsidence based on gated recurrent neural network

作　　者：火天宝 何毅 姚圣 张立峰 张清 HUO Tianbao;HE Yi;YAO Sheng;HANG Lifeng;ZHANG Qing(Faculty of Geomatics,Lanzhou Jiaotong University,Lanzhou 730070,China;National-Local Joint Engineering Research Center of Technologies and Applications for National Geographic State Monitoring,Lanzhou 730070,China;Gansu Provincial Engineering Laboratory for National Geographic State Monitoring,Lanzhou 730070,China)

机构地区：[1]兰州交通大学测绘与地理信息学院,甘肃兰州730070 [2]地理国情监测技术应用国家地方联合工程研究中心,甘肃兰州730070 [3]甘肃省地理国情监测工程实验室,甘肃兰州730070

出　　处：《海洋测绘》2024年第3期47-52,共6页Hydrographic Surveying and Charting

基　　金：国家自然科学基金(42201459);甘肃省教育厅:青年博士基金(2022QB-058);甘肃省自然科学基金(20JR10RA249)。

摘　　要：为防止中国香港国际机场填海造陆引起的地面沉降对基础设施造成破坏,利用永久散射体合成孔径雷达干涉(PS-InSAR)技术,获得中国香港国际机场2016~2020年地面沉降数据,并利用小基线集雷达干涉(SBAS-InSAR)技术以及水准点数据验证;引入门控循环(GRU)神经网络构建堆叠式GRU地面沉降预测模型,对中国香港国际机场未来地面沉降进行时序预测,并与SVM和MLP神经网络进行比较。结果表明:中国香港国际机场2016~2020年地面沉降空间分布不均匀,累计沉降逐渐增加,2020年12月垂直向的累积沉降量已达106 mm。构建的堆叠式GRU神经网络地面沉降方法相比SVM和MLP更准确,2021年7月中国香港国际机场最大累积地面沉降可达111.8 mm。本文提出的地面沉降时序预测模型,可作为一种有效预测地面沉降的方法,为地面沉降早期预警提供关键技术支持。To prevent damage to infrastructure caused by ground subsidence caused by land reclamation at Hong Kong International Airport,China,this paper used the Persistent Scatterer Interferometric Synthetic Aperture Radar(PS-InSAR)technique to obtain ground subsidence data for 2016-2020 at Hong Kong International Airport,China.We used the Small Baseline Subset Interferometric Synthetic Aperture Radar(SBAS-InSAR)technology and level data to verify the results of PS-InSAR.Moreover,Gated Recurrent Unit(GRU)neural network was introduced to construct a stacked GRU ground subsidence prediction model,and the future ground subsidence of Hong Kong International Airport in China was predicted in time-series,and compared with Support Vector Machine(SVM)and Multilayer Perceptron(MLP)neural network.The results showed that the spatial distribution of ground subsidence at Hong Kong International Airport,China,from 2016 to 2020 was uneven,the accumulated subsidence gradually increased,and the accumulated subsidence in the vertical direction in December 2020 reached 106 mm.The constructed stacked GRU neural network ground subsidence method was more accurate than SVM and MLP,and the maximum accumulated ground subsidence at Hong Kong International Airport,China,in July 2021 reached 111.8 mm.The ground subsidence time series prediction model proposed in this paper can be used as an effective method to predict ground subsidence and provide key technical support for early warning of ground subsidence.

关 键 词：永久散射体合成孔径雷达干涉测量 地面沉降 时序预测 门控循环神经网络 填海造陆 

分 类 号：P237[天文地球—摄影测量与遥感]