Deep learning methods for time-dependent reliability analysis of reservoir slopes in spatially variable soils

Abstract

The Three Gorges Reservoir Area (TGRA) is one of the most important landslide-prone regions in China, and rational stability evaluation of reservoir slopes in it is of great significance to design mitigation measures and prevent landslide disasters. It is well recognized that seasonal rainfall and periodic reservoir water level fluctuation are the two major factors influencing the stability of reservoir slopes, and thus the reservoir slope stability may be varying with the external triggering factors. Although geotechnical reliability analysis offers a novel means to quantify slope stability in a probabilistic manner, the previous research focuses more on the time-independent slope reliability analysis, ignoring the effects of time-varying factors. How to evaluate the time-dependent reliability of reservoir slopes accurately and efficiently remains an open question. This study proposes deep learning (DL)-based time-dependent reliability analysis approach, and a practical case adapted from the Bazimen landslide in the TGRA is used for illustration. The predictive performances of the three DL algorithms, namely Convolutional Neural Network (CNN), Long short-term memory (LSTM), and Light gradient boosting machine (LightGBM) are systematically investigated. Results show that the proposed DL-based approach can reasonably portray the variation tendency of the Bazimen landslide time-dependent failure probability, which provides a promising way to rationally evaluate the time-dependent failure probability of reservoir slopes considering the spatial variability of soil properties. Among the three DL algorithms, the CNN performs the best in the Bazimen landslide example.