库岸堆积层滑坡位移的分量响应模式及多因子模型

A component response mode and multi-factor model for accumulation landslide displacement induced by reservoir

为了更加真实地刻画库岸堆积层滑坡位移与影响因素间的显式关系,克服现有的叠加型分量响应模式难以构造位移回归函数的不足,定义了一种新的分解模式——波动比,将位移增量表述为波动比与趋势项增量之积。结合三峡库区王爷庙滑坡实例,采用低通FFT滤波方法得到趋势项位移分量,确定当月降雨量、当月水位变化、前一月波动比值为核心因子组合,通过K-Means算法建立各单因子与波动比值的相关性函数关系,并提出由诱发因素项和状态控制项组合而成的多因子概化模型。同位移实测值及叠加型的ARIMA、SVR、RF、BP预测模型进行对比分析,该方法运算简便、预测效果较好、且位移成分可解释。选取库区其他4个典型滑坡案例综合比较研究,结果进一步验证了该方法的合理性。该模式及模型能够准确反映出滑坡位移随时间推移、多因素影响、状态约束的综合响应,为研究该类型滑坡的演化机制及时空预测提供了新思路。

In order to more realistically describe the explicit relation between the displacement of the reservoir bank accumulation layer landslide and its influencing factors,and overcome the shortcomings of the existing superimposed component response model which is difficult to construct the displacement regression function,a new decomposition mode—fluctuation ratio is defined,while the displacement increment is expressed as the product of the fluctuation ratio and the trend item increment.Wangyemiao landslide,a typical accumulation landslide in the Three Gorges Reservoir area,is investigated as an example.The lowpass FFT filtering method is used to obtain the trend term displacement component.The monthly rainfall,the monthly reservoir water level and the previous month’s fluctuation ratio are determined as the core factor combination.The K-means algorithm is used to establish the correlation function between each single factor and the fluctuation ratio.Then a multi-factor generalized model composed of the evoked factor term and the control term is proposed.Compared with the actual monitoring displacement and the values predicted by several other methods,our model has better prediction effect and simple operation;furthermore,the displacement components can be clearly explained.Four other typical landslides in the Three Gorges Reservoir area are selected for comprehensive comparative study,the results further verify its rationality.This method can accurately reflect the comprehensive response of landslide displacement with time lapse,multi-factor influence and state constraint,which provides a new idea for studying the evolution mechanism and time-space prediction of this type of landslide.