Scholarlink

<p>Cascading dam failure poses a significant challenge to hydraulic infrastructure safety, with breach morphology critically governing resultant flood characteristics. This study develops a machine learning framework to predict key hazard intensity indicators—peak discharge and breach initiation time—under multiple breach scenarios in reservoir cascades. Using a three-reservoir system as a case study, we configured varied breach morphologies and combinations for individual and concurrent failure scenarios. Two-dimensional HEC-RAS simulations identified the governing factors, which subsequently inform ed predictions made by both standard and optimized Random Forest algorithms. The results demonstrate accurate prediction of peak discharge and breach initiation time at the terminal reservoir (R2 &gt; 0.90), with the optimized algorithm showing superior performance in complex multi-breach scenarios (R2 &gt; 0.93). This work confirms the efficacy of machine learning in forecasting cascading dam failure hazards, offering a data-driven methodology for reservoir safety management.</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

Machine Learning-Based Prediction of Hazard Intensity Indicators for Cascading Dam Failures Under Different Breach Combinations

  • 期刊论文

  • 发表日期：

    2026年4月24日

查看全文
  • Xinyi Ma,
  • Xiangyi Ding,
  • Meirong Jia,
  • Xiaojie Tang,
  • Zhongchen Lv

摘要

Cascading dam failure poses a significant challenge to hydraulic infrastructure safety, with breach morphology critically governing resultant flood characteristics. This study develops a machine learning framework to predict key hazard intensity indicators—peak discharge and breach initiation time—under multiple breach scenarios in reservoir cascades. Using a three-reservoir system as a case study, we configured varied breach morphologies and combinations for individual and concurrent failure scenarios. Two-dimensional HEC-RAS simulations identified the governing factors, which subsequently inform ed predictions made by both standard and optimized Random Forest algorithms. The results demonstrate accurate prediction of peak discharge and breach initiation time at the terminal reservoir (R2 > 0.90), with the optimized algorithm showing superior performance in complex multi-breach scenarios (R2 > 0.93). This work confirms the efficacy of machine learning in forecasting cascading dam failure hazards, offering a data-driven methodology for reservoir safety management.