Hydraulic Performance Modeling of Inclined Double Cutoff Walls Beneath Hydraulic Structures Using Optimized Ensemble...

Area of Science:

• Geotechnical Engineering
• Computational Fluid Dynamics
• Machine Learning Applications

Background:

• Hydraulic structures require effective seepage control to prevent uplift forces and ensure stability.
• Inclined double cutoff walls are a common method for managing seepage, but their performance analysis can be complex.
• Traditional analytical methods may not fully capture the intricate relationships between design parameters and seepage control effectiveness.

Purpose of the Study:

• To evaluate the effectiveness of five machine learning models in predicting the performance of inclined double cutoff walls under hydraulic structures.
• To identify the most influential design parameters affecting uplift force, hydraulic gradient, and seepage discharge.
• To develop a practical tool for engineers to optimize cutoff wall design and predict seepage-related outcomes.

Main Methods:

• Utilized a dataset of 630 samples from previous studies, incorporating parameters like relative cutoff wall distance (L/B), inclination angle ratio (θ2/θ1), depth ratio (d2/d1), and relative cutoff depth (d2/D).
• Employed five machine learning models: Random Forest (RF), Adaptive Boosting (AdaBoost), eXtreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), and Categorical Boosting (CatBoost).
• Performed hyperparameter optimization using Bayesian Optimization (BO) with five-fold cross-validation and analyzed feature importance using SHapley Additive exPlanations (SHAP) and Partial Dependence Plots (PDP).

Main Results:

• The CatBoost model demonstrated superior predictive accuracy, achieving R² values above 0.95 for uplift force (U/Uo), 0.93 for hydraulic gradient (iR/iRo), and 0.97 for seepage discharge (q/qo), with low RMSE.
• Feature importance analysis indicated that L/B significantly influences U/Uo and iR/iRo, while d2/D is critical for q/qo.
• Partial Dependence Plots revealed specific relationships: a positive linear trend between L/B and U/Uo, a V-shaped impact of d2/d1 on iR/iRo and q/qo, and complex nonlinear effects of θ2/θ1.

Conclusions:

• Machine learning, particularly CatBoost, offers a robust and accurate approach for analyzing the performance of inclined double cutoff walls.
• The study provides valuable insights into the influence of key design parameters, enabling more informed engineering decisions.
• An interactive Graphical User Interface (GUI) was developed, facilitating the practical application of these predictive models in real-world hydraulic engineering projects.