基于变分模态分解和双向长短时记忆神经网络模型的滑坡位移预测  

An Updated Approach to Predict Landslide Displacement by Combining Variational Modal Decomposition with Bidirectional Long Short-Term Memory Neural Network Model

作　　者：张明岳 李丽敏[1] 温宗周[1] ZHANG Mingyue;LI liming;WEN Zongzhou(College of Electronic Information, Xi'an Polytechnic University , Xi'an 710600, China)

机构地区：[1]西安工程大学电子信息学院,西安710600

出　　处：《山地学报》2021年第6期855-866,共12页Mountain Research

基　　金：陕西省自然科学基础研究计划资助项目(2019JQ-206);陕西省教育厅科学研究资助项目(17JK0346);陕西省技术创新引导专项—科技成果转移与推广计划资助项目(2020CGXNG-009)。

摘　　要：滑坡变形的定量预测是滑坡预警系统中的重要组成部分,滑坡变形受其自身地质条件和众多环境因素共同影响,具有动态、复杂和非线性等特点。针对目前滑坡累积位移—时间序列分析研究中随机性位移无法分解与预测、传统预测模型难以模拟滑坡动态演化特性等问题,本文建立了一种基于组合变分模态分解(Variational Mode Decomposition,VMD)和双向长短时记忆(Bidirectional Long Short-Term Memory,Bi-LSTM)神经网络的复合性滑坡位移动态预测模型。该模型首先利用时间序列分析和VMD将滑坡累积位移分解为趋势项、周期项和随机项位移分量,通过分析滑坡的演化特征和诱发滑坡的关键因素,为各位移分量选择合适的影响因素;然后采用多项式拟合预测趋势项位移、Bi-LSTM神经网络对周期项位移和随机项位移进行多数据驱动的动态预测;最后将各位移分量叠加得到累积位移预测值。以新滩滑坡和八字门滑坡为样本,利用实地观测数据,对本模型的预测精度与工程实用性进行对比评估。实验结果表明,本文提出的模型能较好地表征位移“阶跃式”的变形特征。在预测周期项位移时,Bi-LSTM网络相较于长短时记忆神经网络(Long Short-Term Memory,LSTM)和支持向量机(Support Vector Machine,SVM)具有更高的预测精度,平均相对误差(Mean Relative Error,MRE)分别降低了1.339%和7.817%,均方根误差(Root Mean Square Error,RMSE)分别降低了6.761 mm和27.163 mm。说明该模型不仅预测精度高,且更稳定,可以为滑坡防灾减灾工程的实际应用提供新的思路。Landslide is a complex nonlinear dynamic evolution process of earth surface,governed primarily by local geological conditions with environmental complexity.Quantitative prediction of downslope movement is an indispensable component of landslide early warning system.There have been two technical troubles in predicting slope deformation,among which one is that random displacement could not be decomposed and predicted by numerically resolving of the observed cumulative displacements and time series of a landslide,and another is that the dynamic evolution of a landslide could not be feasibly simulated simply by traditional prediction model.In this study,a dynamic model of displacement prediction was introduced for composite landslides based on a combination of Variational Mode Decomposition(VMD)with Bidirectional Long-Short-Term Memory(Bi-LSTM)neural network.In our proposed model,it used time series analysis and VMD to decompose the observed cumulative displacements of a slope into three components,viz.trend term,periodic term and random term;Then by analyzing the evolution pattern of a landslide and its key factors triggering landslides,appropriate influencing factors were selected for each displacement component;Polynomial fitting was used to predict trend term displacement,and Bi-LSTM neural network to make multi-data-driven dynamic prediction for periodic term as well as random term;A cumulative displacement prediction was obtained by a summation of each component.For accuracy verification and engineering practicability of the model,field observations from two known landslides in China,the Xintan landslide and the Bazimen landslide were collected for comparison and evaluation.The case study verified that the model proposed in this paper can better characterize the“stepwise”deformation characteristics of a slope.As compared with short-term memory neural network and support vector machine,Bi-LSTM neural network had higher prediction accuracy in predicting the periodic term of slope deformation,with the Average

关 键 词：滑坡位移 动态预测 变分模态分解 双向长短时记忆神经网络 新滩滑坡 八字门滑坡 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]