Scholarlink

<p>This study addresses the prediction of debris flow hazards by developing a multi-strategy integrated light gradient boosting machine model (MS-LightGBM). The model, based on the standard LightGBM regression framework, incorporates the sample entropy-based improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN-SE). Additionally, a multi-strategy improved crayfish optimization algorithm (ICOA) is proposed to optimize the tuning of ICEEMDAN-SE and LightGBM. An error correction mechanism is also introduced to enhance prediction accuracy. The model is applied to predict hazard levels using 93 debris flow hazard points in Yajiang County, Sichuan Province, China, based on 13 key influencing factors validated through multicollinearity tests. The MS-LightGBM model was compared with 16 regression models, achieving a root mean square error (RMSE) of 0.04, mean absolute error (MAE) of 0.03, and mean absolute percentage error (MAPE) of 4.67%. These results demonstrate the model’s high efficiency and reliability. Furthermore, in-depth analyses of key features, individual sample results, and interaction effects among features are conducted within the SHapley Additive exPlanations (SHAP) explainability framework, and engineering recommendations are proposed based on these insights.</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

A novel method for predicting debris flow hazard: a multi-strategy fusion approach based on the light gradient boosting machine framework

  • 期刊论文

  • 发表日期：

    2025年3月12日

查看全文
  • Tianlong Wang,
  • Qi Ge,
  • Tianxing Ma,
  • Hao Chen,
  • Rui Luo,
  • Xu Wang,
  • Keying Zhang,
  • Zhaowei Chu,
  • Xiaohui Ni,
  • Hongyue Sun

摘要

This study addresses the prediction of debris flow hazards by developing a multi-strategy integrated light gradient boosting machine model (MS-LightGBM). The model, based on the standard LightGBM regression framework, incorporates the sample entropy-based improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN-SE). Additionally, a multi-strategy improved crayfish optimization algorithm (ICOA) is proposed to optimize the tuning of ICEEMDAN-SE and LightGBM. An error correction mechanism is also introduced to enhance prediction accuracy. The model is applied to predict hazard levels using 93 debris flow hazard points in Yajiang County, Sichuan Province, China, based on 13 key influencing factors validated through multicollinearity tests. The MS-LightGBM model was compared with 16 regression models, achieving a root mean square error (RMSE) of 0.04, mean absolute error (MAE) of 0.03, and mean absolute percentage error (MAPE) of 4.67%. These results demonstrate the model’s high efficiency and reliability. Furthermore, in-depth analyses of key features, individual sample results, and interaction effects among features are conducted within the SHapley Additive exPlanations (SHAP) explainability framework, and engineering recommendations are proposed based on these insights.