滑坡位移的动态概率预测模型

A dynamic probabilistic model for landslide displacement prediction

滑坡位移预测是实现滑坡灾害预报的有效手段,文章利用回声状态网络建立动态预测模型来预测滑坡位移。与传统预测方法相比,动态模型能更好地反映出滑坡演化的动态系统本质;动态模型的建立不需要计算互信息来实现变量选择,也使得预测过程得到简化。考虑到确定性预测的局限性,文章进一步将概率预测的思想引入滑坡位移预测的研究中,提出一种动态预测模型的概率预测方法。所提动态预测模型的预测精度,在白水河和石榴树包两个滑坡位移预测的具体案例中得到了检验;而通过概率预测得到具有明确置信度水平的预测区间,从而也对滑坡状态发展变化趋势给出了更全面的描述。

Landslide displacement prediction is an effective approach for establishing early warning systems of landslide disasters. In this study, echo state networks are employed to build dynamic predictors. As compared to conventional static predictors, dynamic predictors can properly express the dynamic nature of landslide systems. Furthermore, mutual information based the feature selection processes can also be avoided in the modelling processes of dynamic predictors. Considering the limitation of deterministic predictions, the dynamic predictor is further developed into a probabilistic predictor. The effectiveness of the proposed approach is verified in the case studies of the Baishuihe landslide and Shiliushubao landslide. The proposed probabilistic predictor generates prediction intervals with confidence levels, and more comprehensive descriptions for the developing trends of landslides can accordingly be obtained.