基于经验模态分解和LSTM模型的滑坡位移预测  

Landslide displacement prediction based on Empirical Mode Decomposition and Long Short-Term Memory neural network model

作　　者：宋丽伟[1] SONG Liwei(Jilin Electronic Information Vocational and Technical College,Jilin 132012,China)

机构地区：[1]吉林电子信息职业技术学院,吉林吉林132012

出　　处：《人民长江》2020年第5期144-148,共5页Yangtze River

基　　金：吉林省教育厅“十三五”科学技术项目“施工企业基于BIM的工程信息管理应用研究”(JJKH20170156KJ)。

摘　　要：建立高精度的位移预测模型对滑坡的提前预报具有重要意义,然而以往的研究多是选用静态预测模型,无法满足滑坡的动态特性。鉴于此,以三峡库区新滩滑坡为例,选用了近期较为流行的长短时记忆网络(LSTM)模型来对滑坡滑动前的累积位移进行动态预测。首先选用经验模态分解法(EMD)将滑坡累积位移分解成趋势项和周期项,然后利用多项式函数预测趋势项位移;再利用动态LSTM模型预测周期项位移;最后将各分量位移累加得到最终的模型计算值。结果表明:LSTM模型预测结果的均方根误差为17.07 mm,相关性系数达0.999,具有较高的预测精度,为“阶梯状”滑坡位移的预测提供了一种可行的思路。It is very important to establish a high-precision displacement prediction model for the advance prediction of landslides.However,static prediction models were used in most of the previous studies,which can not meet the need of dynamic characteristics of landslides.For this reason,taking Xintan landslide in the Three Gorges Reservoir area as an example,and the Long Short-Term Memory(LSTM)was selected to dynamically predict the cumulative displacement before sliding.Firstly,the Empirical Mode Decomposition method(EMD)was used to decompose the cumulative displacement of landslide into trend term and period term.Then the polynomial function was used to predict the displacement of trend term,and the dynamic Long Short-Term Memory neural network model(LSTM)was used to predict the displacement of period term.Finally,the displacement of each component was accumulated to obtain the final model calculation value.The results showed that the root mean square error of the prediction results of the model was 17.07 mm,the correlation coefficient was 0.999,which had a good prediction accuracy and provided a feasible idea for the prediction of the ladder-like landslide displacement.

关 键 词：滑坡位移预测 时间序列 经验模态分解 长短时记忆网络 新滩滑坡 三峡库区 

分 类 号：P642[天文地球—工程地质学]