三峡库区堆积层滑坡演化机理与变形概率预测研究

The evolution of a reservoir accumulative formation landslide is under the control of seepage field and characterized with multiple physics-based parameters. Different types of uncertainties, which are caused by the randomness and obscureness of landslide system, exist in deterministic forecasting models. Due to the uncertainties associated with landslide systems and ignoring the evolution mechanism, errors are unavoidable and sometimes significant in traditional deterministic methods of landslide deformation forecasting. Thus, mechanism-based probabilistic forecasting is essential for improving the reliability of accumulative formation landslide early warning. In this project, typical accumulative formation landslides in the Three Gorges Reservoir area, are selected as case studies. The lithology and structure of the accumulative formation landslides should be first determined from a field investigation. In this sense, the typical geomechanical model of accumulative formation landslides is developed. And the evolution processes are distinguished from the deformation characteristics. Microstructure, mechanical and hydrologic tests of sliding soil under the combined effect of seepage pressure and drying and wetting cycles have been carried out by means of in-situ and laboratory testing. As a result, the deterioration law of the sliding soil under the combined effect of seepage pressure and drying and wetting cycles have been obtained from the experiment results. In-situ monitoring and numerical simulation will be conducted to obtain time series of multiple physics-based parameters. The modern data mining methods, two-step cluster, Apriori, and decision tree C5.0 algorithms are employed to establish the association rules and quantitative models of landslide evolution processes between hydrological parameters and landslide movement. The association rules extracted are used to study the co-evolution laws. On this basis, the evolution mechanism of the accumulative formation landslide in the Three Gorges Reservoir area is studied. Based on landslide evolution mechanism and quantitative models of landslide evolution processes, a hybrid approach based on bootstrap, extreme learning machine (ELM), and artificial neural network (ANN) methods are proposed to perform probabilistic forecasting of landslide deformation. In summary, this project has the potential for providing fundamental bases and solutions for prediction and prevention of accumulative formation landslides in the Three Gorges Reservoir area.

水库堆积层滑坡具有渗流场主导的多场信息协同演化特征，单纯依据位移时间曲线的数学推演而建立的确定性预测模型，忽视了滑坡的演化机理，忽视了滑坡系统模糊性和随机性引起的诸多不确定性因素，往往难以预测成功，亟需开展水库堆积层滑坡演化机理与变形概率预测研究。本项目选取三峡库区典型堆积层滑坡实例，划分水库堆积层滑坡演化进程；采用原位与室内试验，研究渗透压力与干湿循环作用下滑带结构、渗透与力学特性劣化规律，建立滑带水致劣化模型；通过数值模拟和原位多场监测，建立多场信息时间序列矩阵，开展多场信息数据融合与挖掘，建立多场信息关联规则，研究多场信息协同演化关联特征，揭示水库堆积层滑坡演化机理。建立演化进程阈值模型，实现演化进程判识；基于Bootstrap统计推断理论、采用极限学习机和神经网络算法，构建基于演化进程的变形概率预测模型，为水库堆积层滑坡的预测预报与防控提供理论依据。

本项目依托“三峡库区地质灾害国家野外科学观测研究站”和“长江三峡库区地质灾害研究985优势学科创新平台”，以三峡库区地质灾害大型野外试验场为基地，选取黄土坡滑坡、白水河滑坡、树坪滑坡等为典型案例，通过室内试验、原位试验、现场监测、数值模拟与理论分析相结合的研究手段，系统开展了三峡库区堆积层滑坡演化机理与变形概率预测研究。基于CT扫描、声波测试、SEM扫描电镜，揭示了原状滑带土、巴东组红层等滑坡岩土体物理力学特性劣化规律及变形破坏微细观机理；开展了堆积层滑坡多场信息协同演化物理模型试验，揭示了多场信息转移、迁移规律；结合Copula函数建立降雨、库水位、滑坡位移等多场监测数据协同演化Copula关联特性函数，研究了多场信息协同演化关联特性，建立了VaR风险测度；基于人工智能模型结合不确定性理论，建立了滑坡位移区间预测和概率密度预测智能算法；基于Bootstrap统计推断理论，开展了不完备信息条件下大型复杂水库滑坡可靠度研究；研发了一种滑坡深部大位移自适应监测系统和方法。上述研究成果对于三峡库区堆积层滑坡预测预报与防控具有重要的理论意义和工程应用前景，能够为三峡库区堆积层防治结构的设计提供必要的参数依据，同时为三峡库区堆积层滑坡监测预警系统的建设提供了新的思路和技术方案。