基于时变监测数据的降雨滑坡多目标随机反分析

Multi-objective probabilistic inverse analysis of rainfall-induced landslide based on time-varied data

基于非饱和土流-固耦合理论和贝叶斯理论,建立了边坡的非饱和土流-固耦合随机反演模型,提出了基于马尔科夫链的多目标随机反分析方法,利用位移和孔隙水压力时变监测数据进行多目标和单目标随机反演,并对反演结果进行比较分析。结果表明,多目标随机反分析参数后验分布标准差较单目标随机反分析明显减小。单目标随机反分析只对本目标进行优化,对其他目标的预测误差较大。多目标随机反分析能同时对所有目标进行优化,反演结果对所有目标误差均较小,95%置信区间较单目标明显收窄,采用不同类型监测数据的多目标随机反分析所得结果更为可靠,预测更为准确。

Based on the solid-fluid coupling theory and Bayesian theory, a coupled probabilistic back analysis model is developed for an unsaturated soil slope. A method of multi-objective probabilistic inverse analysis using time-varied data of displacement and pore water pressure is proposed based on Markov chain theory. The results of the multi-objective inverse analysis and single-objective inverse analysis found that the posterior standard deviations of the input parameters obtained by multi-objective inverse analysis are smaller than the single-objective cases. It is also found that the results obtained by single-objective agree with the measurement well but have an unsatisfied prediction on other objectives. Multi-objective inverse analysis could optimize all the objectives simultaneously and the results of inverse analysis can meet well with all the objectives. For the multi-objectives case, 95% uncertainty bounds are narrower than the single-objective cases, and the soil parameters obtained by the multi-objective probabilistic inverse analysis using different types of data are more reasonable, and the prediction using the results of the multi-objective probabilistic inverse analysis is more correct than the single-objective probabilistic inverse analysis results.