Investigation on the reliability calculation method of gravity dam based on CNN-LSTM and Monte Carlo method.

Area of Science:

• Geotechnical Engineering
• Computational Mechanics
• Artificial Intelligence

Background:

• Traditional Monte Carlo (MC) methods for dam stress calculation are computationally intensive and time-consuming.
• Nonlinear dynamic systems, such as those simulating dam stress, present challenges for accurate and efficient analysis.
• Existing methods often struggle with balancing computational accuracy and speed.

Purpose of the Study:

• To improve the accuracy and efficiency of dam stress calculations.
• To develop a novel computational framework for nonlinear dynamic systems.
• To reduce the computational time and workload associated with Monte Carlo simulations.

Main Methods:

• A combined prediction model, DS-FEM-CNN-LSTM, was developed using Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) for nonlinear dam stress simulation.
• The Design of Experiments (DOE) method was employed to optimize sample point design for MC simulations.
• Weight factors and distance to failure surface were utilized as screening criteria within the MC framework.

Main Results:

• The proposed DS-FEM-CNN-LSTM-MC method demonstrated superior performance compared to existing techniques.
• Significant improvements in both computational time consumption and accuracy were achieved.
• The integration of deep learning with MC simulation effectively addressed the limitations of individual methods.

Conclusions:

• The DS-FEM-CNN-LSTM-MC method offers a more efficient and accurate approach for dam stress reliability calculation.
• This hybrid approach provides a robust solution for analyzing complex nonlinear dynamic systems.
• The findings suggest a promising direction for advancing computational methods in geotechnical engineering.