基于变形速率分解的阶跃型滑坡预测——以呷爬滑坡为例  

Prediction of step-type landslides based on deformation rate decomposition——a case study of Gapa landslide

作　　者：邢保印 张炜怡 章广成[1] 张世殊 刘忠绪 曾鑫 郑子涵 XING Baoyin;ZHANG Weiyi;ZHANG Guangcheng;ZHANG Shishu;LIU Zhongxu;ZENG Xin;ZHENG Zihan(Faculty of Engineering,China University of Geosciences(Wuhan),Wuhan,Hubei 430074,China;PowerChina Chengdu Engineering Corporation Limited,Chengdu,Sichuan 610072,China)

机构地区：[1]中国地质大学(武汉)工程学院,湖北武汉430074 [2]中国电建集团成都勘测设计研究院有限公司,四川成都610072

出　　处：《岩石力学与工程学报》2023年第3期685-697,共13页Chinese Journal of Rock Mechanics and Engineering

基　　金：国家自然科学基金资助项目(41931295,41877263)。

摘　　要：阶跃型滑坡位移预测是滑坡变形现状研究与危害评估的重要工作,而阶跃预测的研究多为平稳波动信号,基于滑坡阶跃运动特征的变形速率分解方法提供了非线性变形阶段阶跃滑坡信号分解的新思路。以呷爬滑坡为例,通过差分与离散小波变换(DWT)平滑方法得到变形速率数据,并基于滑坡阶跃运动特征将变形速率数据分解为由外部诱发因素决定的小尺度波动项与内在控制因素决定的大尺度趋势项,其中变形速率趋势项信号通过添加震荡函数的反Logistic函数模型(ILF)预测,并结合曲率极值法判识滑坡变形状态;变形速率波动项信号则运用长短时记忆神经网络(long short-term memory,LSTM)构建非线性映射模型,以降雨、库水位作为诱发输入,趋势项预测结果作为控制输入进行预测。预测结果表明,基于变形速率分解模型针对呷爬滑坡非线性过程数据的预测精度相比传统的位移拟合分解模型更高,外部因素映射能力更强,因此基于变形速率分解是基于阶跃运动机制预测的有效思路,解决了非线性变形阶段的阶跃滑坡预测问题。It is a crucial work in the research of landslide deformation status and hazard assessment to predict the displacement of step-type landslides.Generally,the research of step prediction is stationary fluctuation signal.The deformation rate decomposition method based on the step motion characteristics of landslide provides a new idea of step landslide signal decomposition in nonlinear deformation stage.Taking Gapa landslide as an example,the deformation rate data is obtained by signal differential and smoothing method of discrete wavelet transform(DWT).Based on the step motion characteristics of landslide,the deformation rate data is decomposed into small-scale fluctuation items determined by external inducers and large-scale trend items determined by internal control factors,in which the deformation rate trend signal is predicted by Inverse logistic function model(ILF)with added oscillation function.The deformation state of landslide is judged by curvature extreme value method.Deformation rate fluctuation term signals were predicted by constructing nonlinear mapping models using long short term memory LSTM,with rainfall and water level as evoked inputs,and the prediction results of trend term as control inputs.The prediction results show that the decomposition model based on the rate of deformation is more accurate than the traditional displacement fitting decomposition model for the non-linear process data of the Gapa Landslide,and its ability to map external factors is stronger.Therefore,deformation rate decomposition is an effective idea for prediction based on step motion mechanism,which solve the problem of step landslide prediction in nonlinear deformation stage.

关 键 词：边坡工程 阶跃型滑坡预测 变形速率分解 反Logistic函数模型 长短时记忆神经网络 呷爬滑坡 

分 类 号：P642[天文地球—工程地质学]