冯非凡, 武雪玲, 牛瑞卿, 许石罗. 一种V/S和LSTM结合的滑坡变形分析方法[J]. 武汉大学学报 ( 信息科学版), 2019, 44(5): 784-790. DOI: 10.13203/j.whugis20170218
引用本文: 冯非凡, 武雪玲, 牛瑞卿, 许石罗. 一种V/S和LSTM结合的滑坡变形分析方法[J]. 武汉大学学报 ( 信息科学版), 2019, 44(5): 784-790. DOI: 10.13203/j.whugis20170218
FENG Feifan, WU Xueling, NIU Ruiqing, XU Shiluo. A Landslide Deformation Analysis Method Using V/S and LSTM[J]. Geomatics and Information Science of Wuhan University, 2019, 44(5): 784-790. DOI: 10.13203/j.whugis20170218
Citation: FENG Feifan, WU Xueling, NIU Ruiqing, XU Shiluo. A Landslide Deformation Analysis Method Using V/S and LSTM[J]. Geomatics and Information Science of Wuhan University, 2019, 44(5): 784-790. DOI: 10.13203/j.whugis20170218

一种V/S和LSTM结合的滑坡变形分析方法

A Landslide Deformation Analysis Method Using V/S and LSTM

  • 摘要: 滑坡变形的产生是坡体自身地质条件和外部诱发条件共同作用的结果,滑坡变形定量预测是滑坡监测预警的关键。传统的基于滑坡累计位移-时间曲线分析滑坡变形的方法,忽略了滑坡变形演化的影响因素,难以对滑坡变形进行准确预测。三峡库区滑坡研究多集中在滑坡时空分布特征和滑坡整体稳定性分析方面,亟需开展单体滑坡综合变形分析。以三峡库区白水河滑坡为例,基于滑坡宏观变形和位移监测数据,利用重标方差(rescaled variance statistic,V/S)分析法对滑坡整体和局部变形趋势进行分析,进而构建考虑库水位波动和降雨滞后性影响因素的可有效利用长期依赖信息的长短记忆(long short-term memory,LSTM)神经网络模型,定量预测滑坡位移。研究结果表明,滑坡体属牵引式滑坡,北东部稳定性较差,西部和后缘相对稳定,预测值的均方根误差为8.95 mm,证明该模型是一种高性能的滑坡变形分析方法。

     

    Abstract: Landslide deformation is the result of the combination of the geological conditions and the external induced factors. The quantitative prediction of landslide deformation is the key to landslide monitoring and early warning. The traditional method based on cumulative displacement-time curve of landslide neglects the influence factors of landslide deformation and evolution, it is difficult to predict landslide deformation accurately. Landslide research in the Three Gorges reservoir area is mostly concentrated on the temporal and spatial distribution characteristics of landslides and the stability analysis of landslides. It is urgent to carry out comprehensive deformation analysis of single landslides. Baishuihe landslide is selected as a case study. Based on landslide macroscopic deformation and displacement monitoring data, spatial-temporal deformation trend of the landslide is analyzed using V/S analysis. Then, long short-term memory neural network model is constructed which consider the influence factors of reservoir water level fluctuation and rainfall hysteresis. It can effectively use the long-term dependent information to realize the quantitative prediction of landslide displacement. The results show that the landslide is characterized by traction landslide, deformation tendency gradually increases from southwest to northeast, and the west and trailing edge are relatively stable. The prediction error is 8.95 mm, which proves the model is of great performance to analyze landslide deformation.

     

/

返回文章
返回