Scholarlink

<p>Precise estimation of the longitudinal dispersion coefficient (<i>K</i><sub><i>x</i></sub>​) is essential for modeling pollutant transport in rivers. To address the limitations of existing empirical formulas for <i>K</i><sub><i>x</i></sub>​​ calculation, this study introduces a hybrid machine learning model, SSA-CatBoost, which combines the Sparrow Search Algorithm (SSA) with the CatBoost framework to enhance hyperparameter optimization. Its performance is benchmarked against popular gradient boosting models such as CatBoost, XGBoost and GBDT. Results demonstrate that the newly developed SSA-CatBoost gives better comprehensive prediction performances than other gradient boosting models with RMSE of 878.02 and MAE of 478.88 for dimensionless <i>K</i><sub><i>x</i></sub> and RMSE of 163.15 m<sup>2</sup>/s and MAE of 82.44 m<sup>2</sup>/s for dimensioned <i>K</i><sub><i>x</i></sub>. Furthermore, comparisons with traditional empirical formulas highlight the enhanced precision of SSA-CatBoost. Besides, the uncertainty analysis also suggests that SSA-CatBoost is able to give higher prediction reliability with the uncertainty bandwidth of 584 for dimensionless <i>K</i><sub><i>x</i></sub> and 95 m<sup>2</sup>/s for dimensioned <i>K</i><sub><i>x</i></sub>. These findings establish SSA-CatBoost as a reliable and effective solution for <i>K</i><sub><i>x</i></sub>​ prediction in natural river systems.</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 Hybrid Gradient Boosting Model for Predicting Longitudinal Dispersion Coefficient in Natural Rivers

  • 期刊论文

  • 发表日期：

    2025年2月24日

查看全文
  • Yangyu Deng,
  • Yakun Liu,
  • Di Zhang,
  • Ze Cao

摘要

Precise estimation of the longitudinal dispersion coefficient (Kx​) is essential for modeling pollutant transport in rivers. To address the limitations of existing empirical formulas for Kx​​ calculation, this study introduces a hybrid machine learning model, SSA-CatBoost, which combines the Sparrow Search Algorithm (SSA) with the CatBoost framework to enhance hyperparameter optimization. Its performance is benchmarked against popular gradient boosting models such as CatBoost, XGBoost and GBDT. Results demonstrate that the newly developed SSA-CatBoost gives better comprehensive prediction performances than other gradient boosting models with RMSE of 878.02 and MAE of 478.88 for dimensionless Kx and RMSE of 163.15 m2/s and MAE of 82.44 m2/s for dimensioned Kx. Furthermore, comparisons with traditional empirical formulas highlight the enhanced precision of SSA-CatBoost. Besides, the uncertainty analysis also suggests that SSA-CatBoost is able to give higher prediction reliability with the uncertainty bandwidth of 584 for dimensionless Kx and 95 m2/s for dimensioned Kx. These findings establish SSA-CatBoost as a reliable and effective solution for Kx​ prediction in natural river systems.