Efficient Reliability Analysis of Slopes in Spatially Variable Soils Based on Active-Learning Multivariate Adaptive Regression Spline

Reliability analysis of slopes in spatially variable soils is often faced with a huge computational burden. To effectively reduce the number of numerical models in reliability analysis and alleviate the calculation pressure, this paper proposes an active-learning multivariate adaptive regression spline (MARS) method for reliability analysis of spatially variable slopes. Firstly, Latin hypercube sampling is used to obtain the initial training samples. The active-learning function is then used to update the MARS surrogate model by searching for new training samples around the limit state surface. The failure probability of the slope can be calculated using Monte Carlo simulation (MCS) after obtaining an appropriate surrogate model. A typical slope with spatially variable soils is used as an example to verify the efficiency of the proposed method. The results show that the proposed method not only reduces the quantity of calculations required, but also achieves high accuracy in estimating slope failure probability.

Keywords: Slope reliability, Spatial variability, Random field, Multivariate adaptive regression splines, Active-learning.