Displacement Prediction Of Stepped Landslide Based On HP Filter And XGBoost And Threshold Research

Since the Three Gorges Dam was impounded,due to the combined action of various factors,the number of deformation and damage of landslides in the reservoir area has increased significantly,and the reservoir area has become an area with frequent geological disasters in my country.The step-type landslide is a typical landslide in the reservoir area.Due to the regulation of rainfall and reservoir water level,its deformation changes significantly from May to August each year,and it remains almost unchanged in the rest of the month,so the displacement time curve appears to be stepped.For this type of landslide,the deformation often increases sharply within a few months,and it is difficult to judge when the landslide is damaged,which seriously affects the safety of life and property of people in the adjacent areas.Therefore,carrying out research on the displacement prediction and threshold setting of step-type landslides can provide a scientific basis to avoid major losses caused by landslides.This paper takes the Bazimen landslide and the Baishuihe landslide as the research objects.The main research results are as follows:(1)A prediction model based on HP filtering and extreme gradient boosting algorithm(XGBoost)is proposed to study the displacement prediction of stepped landslides.(1)According to the analysis of displacement time series,HP filter is used to decompose the total landslide displacement into trend term displacement and periodic term displacement.The trend term displacement is predicted by using a polynomial function,and the predicted result differs very little from the actual result.At the same time,compared with the prediction of trend term displacement extracted by moving average method and empirical mode decomposition(EMD),the prediction accuracy of the prediction of trend term displacement extracted by HP filtering is the highest.(2)The periodic term displacement is affected by a variety of factors(rainfall,reservoir water level),and the appropriate inducing factors are selected by means of gray correlation degree.Three prediction models of XGBoost,Support Vector Machine(SVR)and Back Propagation Neural Network(BP-NN)are used to predict the periodic term displacement.The results show that the prediction effect of the XGBoost prediction model is better than that of the SVR prediction model and the BP-NN prediction model.In addition,the XGBoost prediction model in the prediction of obvious changes in displacement,the predicted trend is basically the same as the actual trend,and the prediction error is not much different.Finally,the prediction result of total displacement is obtained by superimposing the prediction results of the trend term displacement and the periodic term displacement,and the error is within an acceptable range.(2)Kernel density estimation,Copula function and Value-at-risk model(Va R)are used to calculate the thresholds for monthly displacement,rainfall and reservoir water level fluctuations for the step-type landslide.(1)Kendall correlation coefficient and Spearman correlation coefficient are used to determine the lag months of monthly displacement variation,rainfall and reservoir water level fluctuations.The results show that the monthly displacement change has the greatest correlation with the rainfall with a lag of one month and the fluctuation of the reservoir water level with a lag of four months.(2)Kernel density estimation is used to fit the respective marginal distribution functions of monthly displacement variation,one-month-lag rainfall,and four-month-lag reservoir water level fluctuations.Then,the Copula function is used to construct a binary Copula joint distribution function of monthly displacement change and one-month lag rainfall,monthly displacement change and four-month lag reservoir water level fluctuation.Finally,Va R is used to inversely solve the binary Copula joint distribution function,the threshold of landslide is obtained.The thresholds of monthly displacement change,one-month-lag rainfall and four-month-lag reservoir water level fluctuations for the Bazimen landslide are 119.4mm,268.5mm,and 9.4m,respectively.The thresholds of monthly displacement change,one-month-lag rainfall and four-month-lag reservoir water level fluctuations for the Baishuihe landslide are 103.9mm,288.5mm,and 9.6m,respectively.(3)In the previous analysis,only the effects of monthly displacement variation and rainfall or reservoir water level fluctuations were considered.For further improvement,considering the comprehensive effects of monthly displacement change,rainfall,and reservoir water level fluctuation,a ternary Copula joint distribution function of monthly displacement change,one-month-lag rainfall and four-month-lag reservoir water level fluctuation is constructed.Finally,the thresholds of monthly displacement change,onemonth-lag rainfall and four-month-lag reservoir water level fluctuations for the Bazimen landslide are 164.5mm,311.6mm,and 9.7m,respectively.The thresholds of monthly displacement change,one-month-lag rainfall and four-month-lag reservoir water level fluctuations for the Baishuihe landslide are 199.9mm,298.3mm,and 10.5m,respectively.In addition,these three thresholds need to be met at the same time.