Time-dependent reliability analysis of unsaturated slopes under rapid drawdown with intelligent surrogate models

Slope stability in reservoirs depends on time-dependent triggering factors such as fluctuations of the groundwater level and precipitation. This paper assesses the stability of reservoir slopes over time, accounting for the uncertainty of the shear strength and hydraulic parameters. An intelligent surrogate model has been developed to reduce the computational effort. The capability of two machine learning algorithms, namely Support Vector Regression and Extreme Gradient Boosting, is considered to obtain the relationship between geomechanical parameters and the factor of safety. The probability of failure of a hypothetical reservoir slope is estimated employing Monte Carlo simulations for different scenarios of drawdown velocity. A sensitivity analysis is conducted to investigate the influence of the geomechanical parameters, regarded as random variables, on the probability of failure. The results revealed that the coefficient of variation in the effective friction angle and the correlation between effective cohesion and friction angle have the highest impact on the probability of failure. The intelligent surrogate model can predict the factor of safety of reservoir slopes under rapid drawdown with high accuracy and enhanced computational efficiency.

Automated summary

This paper evaluates and analyzes the stability of reservoir slopes, develops an intelligent surrogate model, and uses two machine learning algorithms to predict the relationship between geomechanical parameters and the factor of safety. The probability of failure is estimated through Monte Carlo simulations. Sensitivity analysis shows that the coefficient of variation in the effective friction angle and the correlation between effective cohesion and friction angle have the highest impact on the probability of failure.