降雨及库水位共同作用下滑坡阶跃位移临界面的识别方法

A method for identifying critical displacement threshold of landslide step under combined effects of rainfall and reservoir water level

针对目前“阶跃式”滑坡位移预测模型的不足,提出一种表征“阶跃式”滑坡特征的降雨及库水位作用下的阶跃位移预测模型。该模型首先采用西原蠕变模型拟合滑坡趋势项位移,经过自适应遗传算法下的滑动阶跃识别方法与移动平均算法下的阶跃位移提取方法得到阶跃位移;然后针对阶跃位移,分析选取降雨及库水位影响因子,经过粒子群优化算法(PSO)得到最优相关性下的降雨及库水位综合指标;最后对数据集进行LOF(local outlier factor)算法剔除离群值处理与ADASYN(adaptive synthetic)算法过采样处理,经过蝗虫优化算法优化支持向量机(GOA-SVM)进行二分类后实现对滑坡阶跃位移的分类与预测。以“阶跃式”滑坡——白水河滑坡为例,选取监测点ZG118从2003年6月~2009年8月典型阶跃时间段的数据进行研究。预测结果:数据集总体预测准确率为72.73%,阶跃位移数据集预测准确率为100%,未发生阶跃位移数据集预测准确率为66.67%,预测效果良好。当降雨与库水位作用成为“阶跃式”滑坡阶跃发生的主导因素时,该模型为此类滑坡位移预测提供了新的思路与探索。

To overcome the deficiency of the current “step-type” landslide displacement prediction model,a new model is proposed to characterize the features of “step-type” landslides under the combined effects of rainfall and reservoir water level.The model first uses the Nishihara creep model to fit the trend term displacement of the landslide,and then obtains the step displacement through the sliding step identification method based on the adaptive Genetic Algorithm and the step displacement extraction method based on the Moving Average Algorithm.After the step displacement is extracted,the influencing factors of rainfall and reservoir water level are analyzed and selected,and the optimal comprehensive index of rainfall and reservoir water level under the best correlation is obtained utilizing the Particle Swarm Optimization(PSO) algorithm.Finally,the LOF(Local Outlier Factor) Algorithm is utilized to remove outliers of the data set,and the ADASYN(Adaptive Synthetic) Algorithm is utilized for oversampling.After binary classification by the Grasshopper Optimization Algorithm optimized Support Vector Machine(GOA-SVM),the classification and prediction of landslide step displacement are realized.Taking the Baishuihe landslide as an example of “step-type” landslides,the data of monitoring point ZG118 in the typical step period from June 2003 to August 2009 are selected for research.The prediction results show that the overall prediction accuracy of the data set is 72.73%,the prediction accuracy of the step displacement data set is 100%,and the prediction accuracy of the data set without step displacement is 66.67%,It indicates good prediction performance.When rainfall and reservoir water level become the dominant factors of the occurrence of step-type landslide steps,this model provides new ideas and exploration for the prediction of such landslides.