[关键词]
[摘要]
针对蠕变型滑坡变形具有渐变性、阶段性和变异性等特征,应用数据挖掘技术提出基于多变量监测时间序列的滑坡预警方法。采用动态时间规划方法度量边坡变形多测点时间序列间的相似性,确定边坡的时空变形规律;考虑主要影响因素和边坡变形等多变量,利用多变量局部异常系数划分滑坡变形演化阶段;基于统计学方法拟定基于局部异常系数的滑坡预警阈值,提出基于多变量局部异常系数阈值的滑坡预警方法。卧龙寺和新滩滑坡验证算例结果表明,提出的方法对累计位移、变形速率和加速度均较敏感,比已有估算方法更具时效性,能防止因环境因素波动而造成的误判,具有更好稳健性。提出的方法应用于某渠坡滑坡预警,结果表明渠坡现阶段较为稳定。因此,提出的方法可为蠕变型滑坡预警提供参考。
[Key word]
[Abstract]
Creep-type landslides generally have three stages from deformation to failure such as initial deformation,uniform deformation and accelerated deformation.According to the characteristics of deformation stages,a variety of landslide warning methods,based on thresholds such as cumulative displacement,deformation rate,and acceleration being proposed.In recent years,data mining technique has been applied to landslide warning.However,related researches have neglected the mining of the spatial deformation characteristics of multiple measuring points and the evolution laws of multivariable for landslides.Therefore,the data mining technique was applied to propose a landslide warning method based on the threshold of local outlier factors of multivariate monitoring time series.The proposed method can provide a reference for the judgment of the creep-type slope stages and the early warning of landslides. The dynamic time warping method (DTW) was used to calculate the distances among the deformation time series of multiple measuring points of the slope,and the Huygens theorem-based ward standard was employed to determine the clusters of the deformation measuring points.The similarities of the deformation laws of multiple measuring points were identified,and the spatiotemporal deformation evolution law of the slope was analyzed.Considering the multi-variables such as the main influencing factors and the multiple deformation measuring points,the local outlier factors (LOF) of the multivariable time series were calculated,and the abnormality existing in the multivariable change processes was analyzed.Based on the statistical methods,the early warning thresholds of LOFs under different confidence levels were determined,to divide the initial,uniform,and accelerated deformation stages of the landslide.Examples of Wolongsi and Xintan landslides were used to verify the effectiveness of the proposed method,and then applied to the creep-type landslide early warning of a canal slope. Two days before the landslide is the critical point of the accelerated deformation stage of the Wolongsi landslide.The LOF values of the cumulative displacement and the deformation rate of the deformation of the measuring points exceeded the thresholds at 95% and 99% confidence levels before two days and one day of landslide.The Xintan landslide experienced four stages:initial deformation,uniform deformation,acceleration deformation,and sharp acceleration deformation.The landslide entered the accelerated deformation stage in July 1982 and occurred on June 12,1985.The time series of the two measuring points A3 and B3 have high similarities.During the uniform deformation stage,the LOF values of the cumulative displacement in each rainy season had a corresponding increase process.In May 1982,the LOF values of the cumulative displacement exceeded the 95% confidence threshold.The LOF values continued to increase,approached the threshold of 99% confidence in August,and the LOF values of the deformation rate began to be at a higher value.In July 1984,the LOF values of the deformation rate exceeded the 95% confidence threshold many times,and in May 1985,it rose sharply and exceeded the 99% confidence threshold.The LOF values can identify the accelerated deformation stage earlier.The canal section X+740~X+860 of a water transfer project had a deep excavation slope.Three variables such as the rainfall and the cumulative displacements of the two measuring points 805-3-m and 835-3-m were considered.The LOF values of multivariate fluctuate slightly during the initial deformation stage.After entering the stable development stage,the multivariate LOF values fluctuate slightly,much less than the threshold when the confidence level is 95%.This canal slope was in a relatively stable development stage. The DTW based similarity measurement method can better reflect the correlation between the displacement time series of slope deformation measurement points.The clustering of the time series of the deformation measuring points can reasonably divide the measuring points into clusters.These can reflect the spatial similarity and heterogeneity of the change laws of the slope deformation measurement points,and facilitate the identification of the shear characteristics of the slope deformation.The LOF values of the deformation process have significant abrupt changes in the initial and the accelerated deformation stages.The LOF threshold based statistical methods for early warning of landslide is determined to divide the deformation stage and its evolution process can overcome the limitation of human estimation.The LOF value is more sensitive to time than human judgment estimation.The proposed early warning method based on the multivariable LOF is highly reliable and can gain more time for taking-emergency measures.The multivariable LOF values among multi-measuring points and external factors have the higher self-adjustment ability and better robustness.The landslide warning based on the multivariable LOF value thresholds can avoid the misjudgment of the landslide caused by the fluctuation of the measured value of accidental factors.
[中图分类号]
[基金项目]