Calculation of Slope Safety Factor Based on Deep Learning Response Surface

Abstract

The slope safety factor is a crucial indicator for assessing slope stability. However, the current methods for calculating safety factors are predominantly based on the search of limit equilibrium theory and the iteration of finite element methods, leading to overly intricate computational procedures. Considering classical mechanics theory and the definition of slope safety factors, there inevitably exists a certain functional relationship between various slope parameters and their safety factors. Thus, we propose an approach utilizing response surface surrogate functions to express this relationship.We studied two types of slopes: soil slopes and rock slopes. For soil slopes, assumed to be single-layer saturated clay, we considered five parameters: soil density, cohesion, friction angle, slope height, and slope angle. For rock slopes, we considered six parameters: rock density, uniaxial compressive strength of the rock, GSI, mi, slope height, and slope angle.we introduce a data sampling technique based on genetic algorithms to enhance the quality of training data. This approach reduces the uncertainty in fitting outcomes while minimizing the volume of sample data, while still meeting precision requirements and generalizability.To address the demands of this study, we establish a convolutional neural network to approximate the response surface. A comparison is made with response surfaces approximated using FCNN and polynomial methods, revealing superior performance of the convolutional neural network. Following training, the surrogate function derived enables rapid and accurate computation of the slope safety factor.