Prediction Model for Deformation of Concrete Dam Based on Interpretable Component Decomposition and Integration

Area of Science:

• Geotechnical Engineering
• Civil Engineering
• Computational Mechanics

Background:

• Concrete dam deformation monitoring faces challenges with prediction accuracy and interpretability due to complex environmental and temporal factors.
• Existing models struggle to identify key influencing factors and provide reliable predictions for dam safety.
• The need for advanced analytical techniques to ensure the structural integrity of dams is critical.

Purpose of the Study:

• To propose a novel, interpretable dam deformation prediction method for concrete dams.
• To improve the accuracy and reliability of deformation monitoring value predictions.
• To enhance the interpretability of prediction models by quantifying factor contributions.

Main Methods:

• Decomposition of deformation data into temporal, environmental load, and residual components using CEEMDAN.
• Prediction of decomposed components using logarithmic functions, Black-Winged Kite Algorithm (BKA)-optimized BiLSTM networks, and cloud models.
• Integration of component predictions and application of SHAP for factor contribution analysis.

Main Results:

• The proposed method demonstrated superior prediction accuracy and trend tracking ability compared to existing models.
• Significant improvements in the reliability and interpretability of dam deformation predictions were achieved.
• The SHAP method effectively quantified the contribution of input factors, enhancing model transparency.

Conclusions:

• The developed interpretable component decomposition and integration method offers a reliable technique for dam deformation safety monitoring.
• This approach addresses limitations in existing models, providing more accurate and understandable predictions.
• The findings contribute to advancing the safety assessment and management of concrete dams.