多模型融合下的滑坡定量预测方法及优化

Landslide quantitative prediction is significant for landslide monitoring, early warning and assessment. Landslide susceptibility assessment and landslide displacement prediction are two difficult problems of landslide quantitative prediction. In this project, Zigui-Badong section of the Three Gorges is selected as study area. Main line of the research is “choice of conditioning factors - landslide susceptibility assessment - landslide displacement prediction” and multi-model integration method is used to solve the redundancy of assessment factors and non-applicability of prediction models. Firstly, a variety of environment and triggering factors of landslides are extracted from the multi-source data fusion, and key factors are chose using rough sets. Then, genetic algorithm was used to optimize the kernel function parameters of support vector machine, and the optimized support vector machine model is trained and used to map landslide susceptibility quantitatively. Finally, based on the spatial-temporal revolution characteristics of typical landslides in the high susceptibility area and deformation mechanism of rainfall-induced landslides, reservoir-induced landslides, rainfall combined reservoir-induced landslides, the total displacements are divided into its trend and periodic components by means of the wavelet analysis. A curve estimation of historical data is used to predict the trend displacement via the curve fitting of the historical displacement versus time. An optimized support vector machine model is proposed to predict periodic displacement. The research results of the project may provide advanced methodology support for landslide disaster prevention and mitigation.

滑坡定量预测是滑坡监测、预警和评估的关键，包括滑坡易发性评价和滑坡位移预测两个难题。本项目以三峡库区秭归至巴东段长江干流岸坡为研究区域，采用多模型融合思路来处理滑坡这种复杂非线性系统，寻求解决上述难题中影响因子冗余和预测模型不适用等问题的定量方法，以“影响因子遴选—滑坡易发性评价—滑坡位移预测”为研究主线。首先，实施地形、地质和遥感等多源数据融合，提取滑坡孕灾环境和诱发因素的信息，通过粗糙集模型来约简滑坡易发性评价和位移预测的影响因子；其次，用遗传算法优化支持向量机核函数参数，构建进化支持向量机模型来定量评价滑坡易发性；最后，分析高易发区内典型滑坡的时空演化特征，分别研究降雨型、库水型和降雨-库水型滑坡的变形机理，采用小波变换模型将滑坡位移分解为趋势项和周期项，通过历史数据曲线拟合预测趋势项位移，采用进化支持向量机模型预测周期项位移。本项目的研究成果有望为滑坡防灾减灾提供先进的方法学支持。

随着人类工程活动的增强，在修建公路、铁路、水库等过程中，不仅造成严重的环境破坏，而且极易影响岩土体的稳定性，诱发滑坡，从而造成巨大的人员伤亡和财产损失。并且，滑坡灾害近年来发生的频率持显著上升趋势。因此，对滑坡灾害进行定量预测研究显得十分迫切。本项目以三峡库区秭归至巴东段长江干流岸坡为研究区域，采用多模型融合思路定量分析了滑坡这种复杂非线性系统中“影响因子遴选—滑坡易发性评价—滑坡位移预测”等关键问题。首先，基于多源监测数据和多模型融合实现动态提取滑坡诱发因子的技术方法，同时对诱发因子的科学分类和重要性排序；其次，构建了遗传算法优化支持向量机耦合模型评价滑坡易发性，绘制滑坡灾害易发性分区图；最后，采用了小波变换将滑坡特征点位移分解为趋势项和周期项位移，通过历史数据曲线拟合预测趋势项位移，构建进化支持向量机回归预测周期性位移。本项目的研究成果有望为滑坡灾害预警提供重要依据，并为后续滑坡灾害风险评估奠定基础，具有十分重要的学术价值和现实意义。