Scholarlink

<p>The First-Order Reliability Method (FORM) was employed in this study to ascertain the probability of liquefaction. This method used six geotechnical and seismic parameters to calculate the reliability index. The data from the cone penetration test (CPT) was implemented. Further, this study utilised normalized cone penetration resistance (qc1N) and cyclic stress ratio (CSR) as its primary input data to construct five machine learning (ML) models: eXtreme Gradient Boosting (XGBoost), Gradient Boosting Machine (GBM), Random Forest (RF), XGBoost-Support Vector Machine (XGBoost-SVM), and XGBoost-Artificial Rabbits Optimizations (XGBoost-ARO). Ten statistical metrics were employed to evaluate the performance of the model. Alongside, its efficiency was evaluated using four methods: score ranking, objective function criterion (OBJ), Akaike Information Criterion (AIC), and uncertainty analysis. XGBoost-SVM achieved the best results, with R² values of 0.92136 (training) and 0.83523 (testing), the lowest OBJ (0.08206), the lowest AIC (–1336.0 training; − 462.3 testing), and the narrowest testing bandwidth (0.30848). Visual analyses proved that the model possessed superior predictive capabilities. The most dependable model for predicting liquefaction probabilities was the XGBoost-SVM model.</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

CPT-Based Data-Driven Approach for Soil Liquefaction Reliability Using Stacked Ensemble and Hybrid Optimization

  • 期刊论文

  • 发表日期：

    2026年4月13日

查看全文
  • Dhilipkumar B,
  • Pijush Samui

摘要

The First-Order Reliability Method (FORM) was employed in this study to ascertain the probability of liquefaction. This method used six geotechnical and seismic parameters to calculate the reliability index. The data from the cone penetration test (CPT) was implemented. Further, this study utilised normalized cone penetration resistance (qc1N) and cyclic stress ratio (CSR) as its primary input data to construct five machine learning (ML) models: eXtreme Gradient Boosting (XGBoost), Gradient Boosting Machine (GBM), Random Forest (RF), XGBoost-Support Vector Machine (XGBoost-SVM), and XGBoost-Artificial Rabbits Optimizations (XGBoost-ARO). Ten statistical metrics were employed to evaluate the performance of the model. Alongside, its efficiency was evaluated using four methods: score ranking, objective function criterion (OBJ), Akaike Information Criterion (AIC), and uncertainty analysis. XGBoost-SVM achieved the best results, with R² values of 0.92136 (training) and 0.83523 (testing), the lowest OBJ (0.08206), the lowest AIC (–1336.0 training; − 462.3 testing), and the narrowest testing bandwidth (0.30848). Visual analyses proved that the model possessed superior predictive capabilities. The most dependable model for predicting liquefaction probabilities was the XGBoost-SVM model.