一种ICEEMDAN-CNN-SVR滑坡位移组合预测模型  

A Combined ICEEMDAN-CNN-SVR Landslide Displacement Prediction Model

作　　者：石化波 王刚 曾怀恩 SHI Huabo;WANG Gang;ZENG Huaien(China Coal Hubei Geological Survey Foundation Engineering Co.,Ltd,Wuhan 430070,China;China Railway Construction Urban Transportation Development Company Limited,Suzhou 215000,China;National Field Observation and Research Station of Landslides in the Three Gorges Reservoir Area of Yangtze River,Yichang 443002,China;College of Civil Engineering and Architecture,China Three Gorges University,Yichang 443002,China)

机构地区：[1]中煤湖北地质勘察基础工程有限公司,武汉430070 [2]中铁建城建交通发展有限公司,江苏苏州215000 [3]湖北长江三峡滑坡国家野外科学观测研究站,湖北宜昌443002 [4]三峡大学土木与建筑学院,湖北宜昌443002

出　　处：《三峡大学学报（自然科学版）》2025年第1期37-43,共7页Journal of China Three Gorges University：Natural Sciences

基　　金：国家自然科学基金项目(42074005)。

摘　　要：滑坡位移预测是滑坡早期预警系统的重要组成部分,针对位移分解程度与特征选取深入程度不够导致滑坡位移预测精度不高的问题,提出一种ICEEMDAN-CNN-SVR滑坡位移组合预测模型:为了解决位移分解程度不够的问题,该模型首先运用ICEEMDAN分解模型对滑坡位移曲线进行分解,将平滑性较好且具有递增趋势的IMF曲线作为趋势项位移,将其他具有波动趋势的IMF曲线总和重构为周期项位移;为了解决特征选取深入程度不够的问题,针对不同位移特性进行了特征变量选取,通过二维平铺与CNN特征提取得到特征变量更深层次的信息,将提取到的特征信息输入SVR预测模型中实现对趋势项位移与周期项位移的精准预测.以典型堆积层滑坡———八字门滑坡为例,选取ZG110与ZG111监测点2007年1月—2012年9月典型变形阶段水平位移数据进行研究,结果表明:ZG110与ZG111监测点预测评价指标R2,ERMSE,EMAE分别为0.9951、0.9989、5.7489、2.7532,4.5091、1.8529,预测效果良好;将模型预测结果与EEMDCNN-SVR预测模型及CNN-SVR预测模型结果作对比,相较其他预测模型,新模型的预测精度有所提升.Landslide displacement prediction is an important component of early warning systems for landslides.In response to the problem of low accuracy in landslide displacement prediction caused by insufficient depth of displacement decomposition and feature selection,an ICEEMDAN-CNN-SVR landslide displacement combination prediction model is proposed.To solve the problem of insufficient displacement decomposition,the ICEEMDAN decomposition model was first used to decompose the landslide displacement curve.The IMF curve with good smoothness and increasing trend was used as the trend term displacement,and the sum of other IMF curves with fluctuating trend is reconstructed as the periodic term displacement.In order to solve the problem of insufficient depth in feature selection,feature variable selection was carried out for different displacement characteristics.Through two-dimensional tiling and CNN feature extraction,deeper information about the feature variables was obtained.The extracted feature information was input into the SVR prediction model to achieve accurate prediction of trend displacement and periodic displacement.Taking the typical accumulation layer landslide,i.e.,Bazimen landslide,as an example,the horizontal displacement data of ZG110 and ZG111 monitoring points during the typical deformation stage from January 2007 to September 2012 were selected for research.The research results showed that the prediction evaluation indicators,R2,ERMSE,and EMAE of ZG110 and ZG111 monitoring points were 0.9951 and 0.9989,5.7489 and 2.7532,4.5091 and 1.8529,respectively,with good prediction effect.Comparing the model prediction results with the EEMD-CNN-SVR prediction model and CNN-SVR prediction model,the comparison results show that the new model has improved prediction accuracy compared to the other prediction models.

关 键 词：八字门滑坡 ICEEMDAN分解 特征提取 CNN-SVR模型 对比分析 

分 类 号：P642.22[天文地球—工程地质学]