Displacement Prediction And Stability Evaluation Methods Of Reservoir Colluvial Landslides In Three Gorges Reservoir Area

Since the Three Gorges Project's impoundment,the large-based rise and periodic scheduling of the reservoir water level have caused significant changes in the reservoir area's hydrogeological conditions.The deterioration of the physical and mechanical properties of the rock and soil mass is accelerated simultaneously.In this case,many reservoir colluvial landslides(RCLs)appear local or overall deformation or even failure.It is of great practical significance and theoretical value to study the landslide displacement prediction and stability analysis method for effectively avoiding or alleviating the damage loss of these landslides.This paper focused on the critical scientific issue of the displacement prediction and stability evaluation methods of RCLs in the Three Gorges Reservoir(TGR)area.Based on the investigation and monitoring data of some RCLs in the TGR area,the development law and deformation characteristics of this kind of landslide were summarized.According to the actual statistical data,a classification system with strong pertinence and flexibility was constructed.Then,the Baishuihe landslide was selected as the research object.By using the data mining,machine learning,mathematical statistics,and numerical simulation methods,a series of research were systematically carried out,including the analysis method of deformation response law,the displacement interval prediction method,the deformation forecasting method,the back analysis method of physical and mechanical parameters,the time-varying failure probability analysis model.The main achievements of this paper were summarized as follows:(1)By analyzing the survey and monitoring data of many RCLs,we summarized the distribution and development characteristics of this kind of landslide in terms of the elevation,wading degree,slope structure,etc.On this basis,considering the classification characteristics and relationship of landslides' cumulative displacement monitoring curves,the spatial deformation characteristics and the whole deformation evolution process were revealed.Through fully summarizing the development law and deformation characteristics,a classification system more suitable for the RCLs in the TGR area was proposed from the comprehensive perspective of the geological condition,triggered factor,and evolution characteristics.(2)For solving the problems existing in the analysis results obtained through the traditional Apriori algorithm,such as the poor adaptability,low calculation efficiency,and high redundancy.A novel improved Apriori algorithm was proposed.This improved algorithm was built based on the data characteristics of multi-field monitoring information of landslides.Taking the improved algorithm as the core,we proposed a comprehensive association rule analysis method for the dynamic response of landslide deformation under external induced factors.In this method,we summarized the basic selection principle of the data preprocessing methods.A multi parameters analysis method was proposed for the deformation event division's inaccurate results.Based on the multi-field monitoring data of the Baishuihe landslide,the deformation response law and the main control factor were analyzed.Then the feasibility and effectiveness of the proposed comprehensive analysis method were verified.(3)To address the shortcomings existing in the landslide point prediction models,such as the low prediction accuracy of mutational samples and the uncertainties of prediction results,a novel interval prediction method was proposed.This method was implemented based on the summary of the sources of uncertainties in landslide displacement prediction and the guidance of interval prediction idea.Based on the method's interval prediction results,a novel early warning method for landslide disasters was proposed.This early warning method was realized through the comprehensive application of the cumulative displacement tangent angle and reliability analysis methods.Considering the monitoring data of the monitoring site ZG118 as an example,we carried out the displacement interval prediction and early warning of the landslide under the coupling effect of rainfall and reservoir water.In this process,the feasibility,validity,accuracy,and reliability of the proposed methods were comprehensively verified.Some problems such as the input factor selection,model structure optimization,model parameters optimization,and early warning criteria adjustment were also discussed.(4)To solve the problems existing in the traditional back-analysis methods,such as the low efficiency,high cost,and low accuracy,we proposed a novel back-analysis method.This proposed method was realized based on the Geo Studio software and MATLAB platform.The non-invasive finite element analysis method and the Bayesian optimization algorithm were used as the core method.Based on the results of the proposed back-analysis method,a time-varying reliability analysis model of RCLs was established to analyze the landslides' overall and local failure probability under the long-term effect of reservoir water and rainfall.We took the actual groundwater and the displacement monitoring data as the basis.The feasibility,effectiveness,and accuracy of proposed methods were verified through the seepage field and stress-strain field inversions and the time-varying failure probability analysis for the landslide.This study's achievements enhance the applicability and practicability of the current advanced data mining methods in the field of landslide disasters and promote the development of medium-term and short-term prediction methods of RCLs.It also improves the existing physical and mechanical parameter back-analysis and stability evaluation methods and may provide theoretical support and practical reference for the disaster prevention and mitigation of RCLs in the TGR area.