联合时序InSAR技术和CS-Elman神经网络的板子沟地表形变监测与预测模型性能的评估  

Evaluation of surface deformation monitoring and prediction model performance in Banzi gully using combined time-series InSAR technology and CS-Elman neural network

作　　者：陈跨越 王保云 CHEN KuaYue;WANG BaoYun(School of Mathematics,Yunnan Normal University,Kunming 650500,China;Yunnan Key Laboratory of Modern Analytical Mathematics and Applications,Kunming 650500,China)

机构地区：[1]云南师范大学数学学院,昆明650500 [2]云南省现代分析数学及应用重点实验室,昆明650500

出　　处：《地球物理学进展》2024年第3期1003-1016,共14页Progress in Geophysics

基　　金：国家自然科学基金“基于深度迁移学习的遥感影像中泥石流孕灾沟谷识别-以云南省为例”(61966040)资助。

摘　　要：汶川县板子沟受2008年5月12日M_(S)8.0大地震影响,造成沟谷内松散物源堆积,近年来滑坡、泥石流灾害频发.针对板子沟地质结构复杂且缺乏有效监测和预测自然灾害等问题,该研究提出一种联合InSAR技术和布谷鸟搜索算法改进Elman(Cuckoo Search-Elman,CS-Elman)神经网络的预测模型来对板子沟地区进行地表形变监测和预测.首先采用SBAS-InSAR和PS-InSAR技术处理覆盖板子沟的22景C波段Sentinel-1A数据,获取地表形变监测值.其次,利用相关性矩阵分析从高程等12个评价因子得出最优评价因子,从多因子角度结合地表形变监测值构建CS-Elman预测模型.最后,通过对比实验分析CS-Elman模型的合理性和优越性.结果表明:(1)SBAS-InSAR和PS-InSAR技术监测同名点雷达视线(Line of Sight,LOS)向形变速率之间的相关系数R^(2)=0.91,具有较高的相关性,证明了两种技术联合分析的可行性;(2)分别选取训练样本数为198、298、398和498,得到CS-Elman模型的预测值与InSAR技术的监测值之间的最大绝对误差分别为11.314 mm、6.188 mm、3.763 mm和2.191 mm,平均绝对误差(Mean Absolute Error,MAE)、均方误差(Mean Squared Error,MSE)、均方根误差(Root Mean Squared Error,RMSE)和平均绝对百分比误差(Mean Absolute Percentage Error,MAPE)也均为样本数为498时最小,分别为0.895 mm、1.712 mm、1.308 mm和5.55%;(3)随机选取518个样本数据,CS-Elman模型的MAE、MSE、RMSE和MAPE分别为1.206 mm、2.052 mm、1.432 mm和6.09%,各项指标均优于Elman模型,验证了CS算法能够有效提高Elman模型的预测精度;(4)通过与GA-BP、CS-SVM模型的对比,验证了CS-Elman模型在地表形变预测中的精度更高,该方法可作为板子沟长时间形变监测和预测的有效手段.The Banzi gully in Wenchuan County was significantly affected by the M_(S)8.0 earthquake on May 12,2008 resulting in the accumulation of loose materials in the gully.In recent years,landslide and debris flow disasters have occurred frequently.Addressing the complex geological structure of Banzi gully and the lack of effective monitoring and prediction of natural disasters,this study proposes an improved prediction model using a combination of InSAR technology and the Cuckoo Search algorithm to enhance the Elman neural network (CS-Elman) for surface deformation monitoring and prediction in the Banzi gully area.Firstly,SBAS-InSAR and PS-InSAR technologies were employed to process 22 scenes of C-band Sentinel-1A data covering Banzi gully,obtaining surface deformation monitoring values.Then,by analyzing the correlation matrix,the optimal evaluation factors were determined from 12 factors including elevation,and a CS-Elman prediction model was constructed by combining the surface deformation monitoring values from a multi-factor perspective.Finally,the rationality and superiority of the CS-Elman model were analyzed through comparative experiments.Results indicate that:(1) SBAS-InSAR and PS-InSAR technologies show a high correlation coefficient(R^(2)=0.91) between the monitored Line of Sight (LOS)deformation rates,proving the feasibility of combined analysis of the two techniques.(2) Selecting training sample sizes of 198,298,398 and 498 respectively,the maximum absolute errors between the predicted values of the CS-Elman model and the monitoring values of InSAR technology are 11.314 mm,6.188 mm,3.763 mm and2.191 mm.The minimum values of Mean Absolute Error(MAE),Mean Squared Error (MSE),Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE) are obtained when the sample size is 498,which are 0.895 mm,1.712 mm,1.308 mm and 5.55%,respectively.(3) Randomly selecting 518 sample data points,the CS-Elman model yields MAE,MSE,RMSE and MAPE values of 1.206 mm,2.052 mm,1.432 mm and 6.09%,respectively,all of whi

关 键 词：小基线集合成孔径雷达干涉测量 永久散射体合成孔径雷达干涉测量 布谷鸟搜索算法 ELMAN神经网络 形变监测 预测分析 泥石流 自然灾害 

分 类 号：P237[天文地球—摄影测量与遥感]