Reliability Analysis of Slope Stability with Intelligent Surrogate Models: A Case Study in the Three Gorges Reservoir

Since the first impoundment of the Three Gorges Dam in China, the yearly fluctuations of the reservoir water level have reactivated several landslides. In this study, the stability of the Huangtupo landslide, one of the largest in the Three Gorges Reservoir region, is investigated with a probabilistic approach. The inherent uncertainty of the shear strength and the hydraulic conductivity is taken into account with the random variable method. Despite the increasing application of probabilistic methods to geotechnical problems, these are rarely applied to complex case studies because of the large computational effort required. To tackle this challenge, surrogate models based on machine learning are used to predict the factor of safety for different levels of the reservoir in a short time. Four models are trained and tested on the data produced with numerical simulations. The best performing model is selected to estimate the probability of failure of one section of the Huangtupo landslide. The results show that surrogate models based on machine learning can predict the factor of safety with good accuracy and improve the computational efficiency of fully probabilistic methods. This study may encourage the application of probabilistic slope stability analysis to complex landslides.