Scholarlink

<p>Spur-dykes are common man-made hydraulic structures that are widely used in alluvial rivers for river restoration and disaster mitigation. Spur dykes are typically built at an angle or perpendicular to the channel bank to protect against bank erosion. The current study evaluates the performance of GLM (Generalized Linear Models), Stacked Ensemble (SE), Gradient Boosting Machine (GBM), and Deep Learning models to estimate the maximum scour depth around spur dykes using observed data and evaluation indicators. The models utilize the following inputs: initial bed level (BL<sub>i</sub>), final bed level (BL<sub>f</sub>), water depth (Y), velocity (V), and mean size (D50) of 0.26&#xa0;mm, 3.27&#xa0;mm, and 5.10&#xa0;mm, D50 is used to enhance the roughness of spur dykes. The GLM model achieves the highest coefficient of determination (R<sup>2</sup>) of 0.995, a Mean Square Error of 0.011, Kling Gupta Efficiency of 0.98, and a Mean Absolute Error of 0.085, outperforming all other models. The best performance of the GLM compared to SE, GBM, and Deep Learning due to its simplicity, alignment with data characteristics, and robustness against overfitting. This study examines the effect of spur dyke roughness on scour depth and leverages advanced machine learning, including Stacked Ensemble and Deep Learning, rarely applied in this field. Furthermore, parametric and sensitivity analyses are carried out to assess the relative significance of the input parameters, which showed BL<sub>f</sub> and BL<sub>i</sub> are the primary contributing factors influencing scour depth.</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

Comparison of Stacking, Boosting, Generalized Linear, and Neural Network Models for Estimating Scour Depth Around Spur-Dykes

  • 期刊论文

  • 发表日期：

    2025年2月25日

查看全文
  • Amit Kumar,
  • A. K. Prabhakar

摘要

Spur-dykes are common man-made hydraulic structures that are widely used in alluvial rivers for river restoration and disaster mitigation. Spur dykes are typically built at an angle or perpendicular to the channel bank to protect against bank erosion. The current study evaluates the performance of GLM (Generalized Linear Models), Stacked Ensemble (SE), Gradient Boosting Machine (GBM), and Deep Learning models to estimate the maximum scour depth around spur dykes using observed data and evaluation indicators. The models utilize the following inputs: initial bed level (BLi), final bed level (BLf), water depth (Y), velocity (V), and mean size (D50) of 0.26 mm, 3.27 mm, and 5.10 mm, D50 is used to enhance the roughness of spur dykes. The GLM model achieves the highest coefficient of determination (R2) of 0.995, a Mean Square Error of 0.011, Kling Gupta Efficiency of 0.98, and a Mean Absolute Error of 0.085, outperforming all other models. The best performance of the GLM compared to SE, GBM, and Deep Learning due to its simplicity, alignment with data characteristics, and robustness against overfitting. This study examines the effect of spur dyke roughness on scour depth and leverages advanced machine learning, including Stacked Ensemble and Deep Learning, rarely applied in this field. Furthermore, parametric and sensitivity analyses are carried out to assess the relative significance of the input parameters, which showed BLf and BLi are the primary contributing factors influencing scour depth.