Efficient time-variant reliability analysis of Bazimen landslide in the Three Gorges Reservoir Area using XGBoost and LightGBM algorithms

Abstract

The Three Gorges Reservoir Area (TGRA) is one of the most important areas for landslide prevention and mitigation in China. Rational reliability analysis of reservoir slope stability is a significant prerequisite for designing mitigation measures and preventing landslide disasters. Due to the influences of periodic reservoir water level fluctuation and seasonal rainfall, the reservoir slope reliability may be varying with the external environment. Although geotechnical reliability analysis has been widely applied, how to evaluate the time-variant reliability accurately and efficiently is still a challenging task. This study develops an efficient time-variant reliability analysis approach by integrating the advanced machine learning algorithms of extreme gradient boosting (XGBoost) and light gradient boosting machine (LightGBM). The proposed approach is applied to a practical case adapted from Bazimen landslide in the TGRA. The performances of XGBoost and LightGBM in the evaluation of Bazimen landslide time-variant failure probability are systematically explored. Results show that the proposed approach can evaluate the time-variant failure probability of Bazimen landslide accurately and efficiently, addressing the prohibitive computational cost of conducting a large number of deterministic analyses at each time instant repeatedly. This greatly facilitates the acquisition of time-variant failure probability for geotechnical engineers in practical applications.