Scholarlink

<p>Based on 393 cases, this study developed an enhanced light gradient boosting machine (KOA–LightGBM) model integrated with the Kepler optimization algorithm (KOA), employing SHAP method to improve interpretability. Key factors including slope height, angle, unit weight, cohesion, internal friction angle, and pore water pressure ratio were determined to evaluate the slope stability under the combined effect of geometric, geotechnical, and hydraulic conditions. Five metrics systematically compared pre-/post-optimization performance of SVM, DT, CatBoost, and LightGBM models, indicating KOA substantially improved the classification capability of all models, with optimized model accuracy not lower than 0.80. Among them, the KOA–LightGBM model performs best, with the accuracy and AUC values of 0.91 and 0.97, respectively, a logarithmic loss of only 0.28, a runtime not exceeding 7.5&#xa0;s, and run-to-run variations within 0.1&#xa0;s. The influence mechanism of each feature on the model output is revealed by the SHAP method, which identifies cohesion, internal friction angle, and unit weight as having positive correlations with the prediction results, whereas slope height, slope angle, and pore water pressure ratio are found to be negatively correlated. Among them, cohesion has the most significant effect on the prediction results of the KOA–LightGBM model, followed by internal friction angle, slope angle, slope height, unit weight and pore water pressure ratio.</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 intelligent evaluation method for slope stability: improved light gradient boosting model based on the Kepler optimization algorithm

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月16日

查看全文
  • Chunyu Chen,
  • Wensong Wang,
  • Ran Tang,
  • Binbin Zheng

摘要

Based on 393 cases, this study developed an enhanced light gradient boosting machine (KOA–LightGBM) model integrated with the Kepler optimization algorithm (KOA), employing SHAP method to improve interpretability. Key factors including slope height, angle, unit weight, cohesion, internal friction angle, and pore water pressure ratio were determined to evaluate the slope stability under the combined effect of geometric, geotechnical, and hydraulic conditions. Five metrics systematically compared pre-/post-optimization performance of SVM, DT, CatBoost, and LightGBM models, indicating KOA substantially improved the classification capability of all models, with optimized model accuracy not lower than 0.80. Among them, the KOA–LightGBM model performs best, with the accuracy and AUC values of 0.91 and 0.97, respectively, a logarithmic loss of only 0.28, a runtime not exceeding 7.5 s, and run-to-run variations within 0.1 s. The influence mechanism of each feature on the model output is revealed by the SHAP method, which identifies cohesion, internal friction angle, and unit weight as having positive correlations with the prediction results, whereas slope height, slope angle, and pore water pressure ratio are found to be negatively correlated. Among them, cohesion has the most significant effect on the prediction results of the KOA–LightGBM model, followed by internal friction angle, slope angle, slope height, unit weight and pore water pressure ratio.