Application of LS-SVM Method in Probabilistic Stability Analysis of Saturated Soil Slopes-PPT Presentation

PPT Presentation

Pdf Summary

This study evaluates least-squares support vector machines (LS-SVM) as a surrogate model for probabilistic stability analysis of saturated, multilayer soil slopes. Because slopes with similar factors of safety (FS) can have different probabilities of failure (PF), the research assesses both measures and compares LS-SVM predictions with limit equilibrium method (LEM) results.<br /><br />The model uses cohesion, internal friction angle, and unit weight as input parameters. LEM analyses of circular slip surfaces provide training data, and Latin Hypercube Sampling (LHS) is used to represent parameter uncertainty. The study examines the effects of uniform versus normal distributions and training-data quantity. LS-SVM models trained on 200 samples are compared with direct Monte Carlo simulation using 30,000 analyses.<br /><br />Results show that LS-SVM closely approximates LEM-based PF and FS while requiring far fewer samples. For the Spencer method, direct simulation gave a PF of 15.95% and FS of 1.15, compared with 15.38% and 1.194 for LS-SVM. For simplified Bishop, the corresponding results were 31.21% and 1.069 versus 30.73% and 1.106.<br /><br />The authors conclude that LS-SVM can reduce computational time and cost while maintaining useful predictive accuracy. Its performance depends on the quality and quantity of training data and appropriate hyperparameter selection. Uniformly distributed cohesion and friction-angle inputs improved the reported recognition of FS and PF. Accurate assessment also requires including all soil layers in the probabilistic model.

Keywords

least-squares support vector machines

LS-SVM

probabilistic stability analysis

saturated soil slopes

multilayer soil slopes

factor of safety

probability of failure

limit equilibrium method

Latin Hypercube Sampling

Monte Carlo simulation