Scholarlink

<p>Time-varying reliability of reservoir slopes subjected to water level drawdown was primarily evaluated using static approaches, in which drawdown rate uncertainty and its influence were often neglected. This study proposed a time-varying reliability assessment framework based on an NGBoost surrogate model with interpretability analysis. A training dataset was generated through numerical simulations, and NGBoost was employed to construct an efficient surrogate model for analyzing the time-varying reliability of unsaturated reservoir slopes considering uncertainties in drawdown rate and soil parameters. The results showed that neglecting drawdown rate uncertainty led to a systematic overestimation of time-varying failure probability, and the magnitude of this effect depended strongly on the drawdown rate. At lower drawdown rates, the time-varying failure probability exhibited a slight increase with increasing the coefficient of variation of the drawdown rate, whereas at higher drawdown rates, the failure probability decreased as the coefficient of variation increased. SHAP analysis revealed the dominant influence of drawdown rate and soil parameters on slope stability, with the drawdown rate exhibiting a pronounced nonlinear effect. The results indicated that appropriate consideration of drawdown rate uncertainty was essential for accurately evaluating the time-varying response of reservoir slopes in stability and reliability analyses.</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 Unsaturated Slopes Using Interpretable Machine Learning Under Soil Parameter and Drawdown Rate Uncertainties

  • 期刊论文

  • 发表日期：

    2026年2月16日

查看全文
  • Jingnian Ran,
  • Lei Sheng,
  • Fei Liu,
  • Yong Fu,
  • Jiang Tao Yi,
  • Xiao Han,
  • Ye Tian

摘要

Time-varying reliability of reservoir slopes subjected to water level drawdown was primarily evaluated using static approaches, in which drawdown rate uncertainty and its influence were often neglected. This study proposed a time-varying reliability assessment framework based on an NGBoost surrogate model with interpretability analysis. A training dataset was generated through numerical simulations, and NGBoost was employed to construct an efficient surrogate model for analyzing the time-varying reliability of unsaturated reservoir slopes considering uncertainties in drawdown rate and soil parameters. The results showed that neglecting drawdown rate uncertainty led to a systematic overestimation of time-varying failure probability, and the magnitude of this effect depended strongly on the drawdown rate. At lower drawdown rates, the time-varying failure probability exhibited a slight increase with increasing the coefficient of variation of the drawdown rate, whereas at higher drawdown rates, the failure probability decreased as the coefficient of variation increased. SHAP analysis revealed the dominant influence of drawdown rate and soil parameters on slope stability, with the drawdown rate exhibiting a pronounced nonlinear effect. The results indicated that appropriate consideration of drawdown rate uncertainty was essential for accurately evaluating the time-varying response of reservoir slopes in stability and reliability analyses.