Scholarlink

<p>This study presents a hybrid machine learning framework for stability assessment of anchored sheet pile walls under uncertainty of soil. Monte Carlo Simulation was employed to generate 1,000 stochastic realizations by incorporating variability in the angle of internal friction, unit weight, and submerged unit weight of sandy soil. Factors of safety against sliding and overturning were computed for each realization to establish a comprehensive dataset. Extreme Gradient Boosting models were optimized using Particle Swarm Optimization, Tiki-Taka Algorithm, Whale Optimization Algorithm, and Reptile Search Algorithm to predict stability responses. Model performance was assessed using statistical error metrics and comparative analyses. The model optimized with Reptile Search Algorithm achieved the highest predictive accuracy for both sliding and overturning stability, surpassing the models optimized with the other metaheuristic algorithms. Explainable artificial intelligence analysis based on SHapley Additive exPlanations was performed to quantify the contribution of input variables and improve model interpretability. The analysis identified the angle of internal friction as the most influential parameter, followed by soil unit weight. The developed framework provided accurate, computationally efficient, and interpretable predictions, demonstrating its applicability for reliability-based assessment and design of anchored sheet pile wall systems under uncertain geotechnical conditions.</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

An Explainable Hybrid XGBoost Framework for Stability of Anchored Sheet Pile Walls

  • 期刊论文

  • 发表日期：

    2026年7月23日

查看全文
  • Sujeet Suman,
  • Md Nadir,
  • Md Nisar Ahmed,
  • Niraj Kumar Nirala

摘要

This study presents a hybrid machine learning framework for stability assessment of anchored sheet pile walls under uncertainty of soil. Monte Carlo Simulation was employed to generate 1,000 stochastic realizations by incorporating variability in the angle of internal friction, unit weight, and submerged unit weight of sandy soil. Factors of safety against sliding and overturning were computed for each realization to establish a comprehensive dataset. Extreme Gradient Boosting models were optimized using Particle Swarm Optimization, Tiki-Taka Algorithm, Whale Optimization Algorithm, and Reptile Search Algorithm to predict stability responses. Model performance was assessed using statistical error metrics and comparative analyses. The model optimized with Reptile Search Algorithm achieved the highest predictive accuracy for both sliding and overturning stability, surpassing the models optimized with the other metaheuristic algorithms. Explainable artificial intelligence analysis based on SHapley Additive exPlanations was performed to quantify the contribution of input variables and improve model interpretability. The analysis identified the angle of internal friction as the most influential parameter, followed by soil unit weight. The developed framework provided accurate, computationally efficient, and interpretable predictions, demonstrating its applicability for reliability-based assessment and design of anchored sheet pile wall systems under uncertain geotechnical conditions.