Scholarlink

<p>A time-variant reliability framework was developed to evaluate slope stability under reservoir drawdown, considering uncertainties in soil parameters and drawdown rate. Training samples were generated using a coupled seepage–stability model, and the factor of safety at discrete time steps was predicted by a hybrid surrogate combining bidirectional long short-term memory, attention mechanism, and adaptive boosting. Failure probability was estimated using Monte Carlo simulation based on surrogate predictions. The model reproduced numerical results with high accuracy, with coefficients of determination up to 0.999 and consistently low prediction errors, while reducing computational time from 15.43&#xa0;h to 15.6&#xa0;min for 2000 samples. The failure probability showed clear dependence on drawdown rate and its variability. Neglecting drawdown-rate uncertainty led to overestimation of failure probability. Under slow drawdown, increasing variability slightly increased failure probability, whereas under rapid drawdown, it reduced failure probability. Interpretability analysis based on Shapley values showed that the effective internal friction angle dominated slope stability, while the drawdown rate mainly controlled short-term transient response and exhibited a nonlinear threshold effect related to pore-pressure dissipation.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Time-Variant Reliability Analysis of Reservoir Slopes Considering Drawdown–Rate Uncertainty Using a Hybrid BiLSTM–Attention–AdaBoost Model

  • 期刊论文

  • 发表日期：

    2026年5月25日

查看全文
  • Jingnian Ran,
  • Lei Sheng,
  • Yu Yang,
  • Ye Tian,
  • Siyu Li,
  • Jiangtao Yi

摘要

A time-variant reliability framework was developed to evaluate slope stability under reservoir drawdown, considering uncertainties in soil parameters and drawdown rate. Training samples were generated using a coupled seepage–stability model, and the factor of safety at discrete time steps was predicted by a hybrid surrogate combining bidirectional long short-term memory, attention mechanism, and adaptive boosting. Failure probability was estimated using Monte Carlo simulation based on surrogate predictions. The model reproduced numerical results with high accuracy, with coefficients of determination up to 0.999 and consistently low prediction errors, while reducing computational time from 15.43 h to 15.6 min for 2000 samples. The failure probability showed clear dependence on drawdown rate and its variability. Neglecting drawdown-rate uncertainty led to overestimation of failure probability. Under slow drawdown, increasing variability slightly increased failure probability, whereas under rapid drawdown, it reduced failure probability. Interpretability analysis based on Shapley values showed that the effective internal friction angle dominated slope stability, while the drawdown rate mainly controlled short-term transient response and exhibited a nonlinear threshold effect related to pore-pressure dissipation.