【摘要】 "阶跃式"滑坡在复杂多变的地质环境作用下呈现突变与稳定交替、不平衡的演化形态,根据此特点提出了边坡突变与稳定分类的平衡集成树模型,建立高维地质环境影响因子与致变之间的关联,并应用于三峡库区26个具有"阶跃"特征的滑坡。考虑到直接使用分类器难以捕捉有效信息,模型利用了合成边界少数类过采样原理适当提升突变样本比率,再分别组合随机森林和梯度提升树进行优化和训练,并对测试集和预测集进行评估校核。同数据平衡前后的不同模型进行对比实验,平衡集成树模型能够有效提高突变预测的整体精度,并量化得出了所有地质环境因子的特征重要性指标,最终应用于样本外的王爷庙滑坡结果表明,模型取得了较高的预报水平。该方法能够实现突变的有效预测,并发现核心影响因子及其数据分布特征,为研究"阶跃式"滑坡机理和预警提供了新的思路。更多还原

【Abstract】 Step-like landslides show abrupt change,stable alternation and unbalanced evolution under the complex and changeable geological environment. Based on this feature,a balanced decision tree ensemble model of slope catastrophe and stable classification is proposed to establish a correlation between highdimensional geological environmental impact factors and mutations,and it is applied to 26 typical step-like landslides in the Three Gorges Reservoir Area. Considering that it is difficult to capture effective information by directly using the classifier,the model uses the synthetic borderline minority over-sampling principle to appropriately increase the mutation sample ratio,and then combines the random forest and the gradient boosted decision tree for optimization and training,and evaluates and verifies the test set and prediction set. Compared with different models before and after data balance,the balanced decision tree ensemble model can effectively improve the overall accuracy of the catastrophe prediction,also the feature importance index of all geological environmental factors is obtained. Finally,the results applied to the Wangyemiao Landslide show that the optimization model has achieved a higher level of forecasting. This method can achieve effective catastrophe prediction,and finds the core influence factors and their data distribution characteristics,which provides the ideas for studying the step-like landslide mechanism and early warning.更多还原