Scholarlink

<p>Bed scour around spur dikes in meandering channels presents a significant hydraulic challenge, compromising riverbank stability and the longevity of river training structures. Conventional empirical and numerical methods often fail to capture the complex, nonlinear sediment-flow interactions inherent to curved channels. Recently, Artificial Intelligence (AI) models have demonstrated considerable promise in addressing these complexities. This study evaluates the predictive capabilities of several AI techniques, including Categorical Boosting (CatBoost) with Particle Swarm Optimization (PSO), Extreme Gradient Boosting (XGBoost) with PSO, Random Forest (RF), and Support Vector Regression (SVR) with PSO, for estimating scour depth in meandering channels. Experimental data were collected from laboratory flume studies, incorporating variables such as spur dike location (Ld), dike permeability (P), channel sinuosity (S), and flow conditions (Fr). The dataset was divided into 70% for training, 15% for testing, and 15% for validation. Model performance was assessed using the coefficient of determination (R²) and root mean square error (RMSE). To enhance interpretability, SHAP analysis was conducted to determine the influence of each parameter on scour depth. The CatBoost-PSO model demonstrated superior performance, achieving R² values of 0.992 in training and 0.985 in testing. The results of the statistical tests (ANOVA and T-test) confirmed the superior performance of the CatBoost-PSO model, with P-values of 0.035 (ANOVA) and 0.041 (T-test). Furthermore, the RMSE value of the CatBoost-PSO in the training and testing phase was 0.00059 and 0.00096, respectively. These results were corroborated by ANOVA, T-tests, and Taylor’s diagram. SHAP analysis indicated that the spur dike location exerted the greatest influence on scour depth, a finding further supported by scatterplot pairs. Overall, the proposed framework demonstrates the robustness of AI-based approaches, with CatBoost-PSO identified as the most reliable model. The results suggest that the strategic placement of spur dikes is an effective strategy for mitigating scour in meandering channels.</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

Artificial intelligence in river dynamics: predicting bed scour in meandering channels with curved spur dike

  • 期刊论文

  • 发表日期：

    2026年3月19日

查看全文
  • Zeeshan Akbar,
  • Ghufran Ahmed Pasha

摘要

Bed scour around spur dikes in meandering channels presents a significant hydraulic challenge, compromising riverbank stability and the longevity of river training structures. Conventional empirical and numerical methods often fail to capture the complex, nonlinear sediment-flow interactions inherent to curved channels. Recently, Artificial Intelligence (AI) models have demonstrated considerable promise in addressing these complexities. This study evaluates the predictive capabilities of several AI techniques, including Categorical Boosting (CatBoost) with Particle Swarm Optimization (PSO), Extreme Gradient Boosting (XGBoost) with PSO, Random Forest (RF), and Support Vector Regression (SVR) with PSO, for estimating scour depth in meandering channels. Experimental data were collected from laboratory flume studies, incorporating variables such as spur dike location (Ld), dike permeability (P), channel sinuosity (S), and flow conditions (Fr). The dataset was divided into 70% for training, 15% for testing, and 15% for validation. Model performance was assessed using the coefficient of determination (R²) and root mean square error (RMSE). To enhance interpretability, SHAP analysis was conducted to determine the influence of each parameter on scour depth. The CatBoost-PSO model demonstrated superior performance, achieving R² values of 0.992 in training and 0.985 in testing. The results of the statistical tests (ANOVA and T-test) confirmed the superior performance of the CatBoost-PSO model, with P-values of 0.035 (ANOVA) and 0.041 (T-test). Furthermore, the RMSE value of the CatBoost-PSO in the training and testing phase was 0.00059 and 0.00096, respectively. These results were corroborated by ANOVA, T-tests, and Taylor’s diagram. SHAP analysis indicated that the spur dike location exerted the greatest influence on scour depth, a finding further supported by scatterplot pairs. Overall, the proposed framework demonstrates the robustness of AI-based approaches, with CatBoost-PSO identified as the most reliable model. The results suggest that the strategic placement of spur dikes is an effective strategy for mitigating scour in meandering channels.