三峡库区堆积层滑坡间歇性活动预测:以白水河滑坡为例

Intermittent Movement Prediction of Colluvial Landslide in the Three Gorges Reservoir: A Case Study of Baishuihe Landslide

三峡库区堆积层滑坡在季节性降雨和库水位周期波动的影响下呈现间歇性活动特征,滑坡活动强度与诱发因素作用强度和时间关系密切。以三峡库区白水河滑坡为例,分析了堆积层滑坡间歇性活动特征和诱发因素,发现降雨和库水位下降是滑坡变形的主要诱发因素。根据滑坡时序曲线特征,将滑坡累积位移分解为趋势项位移和周期项位移,采用多项式拟合的方法来预测趋势项位移,利用长短期记忆神经网络模型来预测周期项位移,并与极限学习机模型、广义回归神经网络模型的预测结果进行了对比分析,发现长短期记忆神经网络模型预测滑坡间歇性活动精度更高。

Colluvial landslides affected by the effect of seasonal rainfall and reservoir water in the Three Gorges Reservoir often behave intermittent movement characteristics. Landslide movement intensity is related to inducing factor intensity and action time. In the paper, Baishuihe landslide in the reservoir was chosen to analyze the characteristics of the intermittent movement and the trigger factors. Rainfall and drawdown of the reservoir water level are found to be major trigger factors. Based on characteristics of time series data, the cumulative displacement is divided into a trend term and a periodic term. A polynomial model was used to predict the trend term and a long short-term memory neural network was used to predict the periodic term. In order to compare prediction results, extreme learning machine and generalized regression neural network were chosen to predict the periodic term. The results show that long short-term memory neural network has higher accuracy in prediction of landslide intermittent movement.