Machine-learning ensembled probabilistic methods for time-dependent reliability analysis of reservoir slopes under...

Area of Science:

• Geotechnical Engineering
• Computational Mechanics
• Machine Learning Applications

Background:

• Probabilistic reservoir slope stability analysis is computationally intensive.
• Existing machine learning (ML) models offer point-value predictions, neglecting model uncertainties.
• Ensembling ML models for improved accuracy and uncertainty quantification remains a challenge.

Purpose of the Study:

• To propose ML ensembled probabilistic methods for time-dependent reliability analysis of reservoir slopes.
• To address computational inefficiency and uncertainty quantification in slope stability analysis.
• To evaluate the performance of ensemble models against individual ML models.

Main Methods:

• Utilized Support Vector Machine (SVM) and Back Propagation-based Neural Network (BPNN) for safety factor (FS) evaluation.
• Employed Bayesian Model Averaging (BMA) to ensemble individual ML model predictions.
• Applied the methods to a practical slope in the Three Gorges Reservoir Area for time-dependent reliability analysis.

Main Results:

• The ensemble surrogate model demonstrated superior predictive performance for slope failure probability compared to individual ML models.
• Ensemble models effectively combined the strengths of different ML algorithms.
• The proposed approach provided probabilistic forecasts, enabling reasonable quantification of predictive uncertainty.

Conclusions:

• ML ensembled probabilistic methods, particularly using BMA, enhance the accuracy of time-dependent reservoir slope failure probability estimation.
• The ensemble approach successfully quantifies predictive uncertainties, offering more reliable risk assessments.
• This methodology provides a more robust framework for reservoir slope reliability analysis.