Ensemble neural network models for stability prediction and optimization of hydraulic structures considering uplift...

Area of Science:

• Geotechnical Engineering
• Computational Fluid Dynamics
• Artificial Intelligence

Background:

• Hydraulic structures require stable designs to prevent failure.
• Traditional methods for analyzing seepage and uplift pressure can be limited.
• Optimizing cutoff wall configurations is crucial for structural integrity.

Purpose of the Study:

• To develop a novel ensemble modeling approach for optimizing hydraulic structure stability.
• To investigate the impact of cutoff wall position and inclination on seepage parameters.
• To create a predictive optimization framework integrating artificial intelligence and numerical simulations.

Main Methods:

• Integration of artificial neural networks (Feed-Forward Neural Network, XGBoost, SVM) with finite element analysis (Geostudio SEEP/W).
• Utilizing a Genetic Algorithm (GA) for predictive optimization of cutoff wall designs.
• Performing numerical simulations to analyze seepage patterns and hydraulic parameters.

Main Results:

• Optimal cutoff wall inclination angle of 165° minimizes uplift pressure and exit gradient across all positions.
• Optimal angles for seepage discharge vary from 60° to 120°, increasing with downstream position.
• The ensemble model achieved high predictive accuracy (R-squared: 0.99 for uplift pressure, 0.94 for seepage discharge, 0.97 for exit gradient).

Conclusions:

• The proposed ensemble modeling framework significantly enhances the prediction and optimization of hydraulic structure stability.
• The integrated approach offers a robust alternative to traditional analytical methods, especially for complex designs.
• Findings provide valuable insights for improving the safety and efficiency of hydraulic infrastructure design.