A dynamic evaluation method for slope safety with monitoring information based on a hybrid intelligence algorithm

The variability of soil or rock parameters based on poor information results in an uncertain change in slope failure probability. To solve the negative impact of slope reliability analysis introduced by variability of soil or rock parameters, this article proposes a hybrid intelligence algorithm for analyzing the slope failure probability with Bayes theorem, integrating a one-dimensional convolutional neural network (1D-CNN), and executing Markov chain Monte Carlo (MCMC) simulation with multidepth displacements. We calculate the failure probability through 1D-CNN prediction. The 1D-CNN learns from data generated by numerical models to directly relate input variables to output variables toward obtaining an accurate model of the response surface. The resulting soil or rock parameters are described as probability distributions, which are updated by MCMC simulations based on the multidepth displacement time series. A case study with monitoring displacement along the slope depth is presented to illustrate the effectiveness of our proposed methodology. We find that our method can update the failure probability value and identify the location of potential sliding surfaces with the time series of multidepth displacement. We also find that incorporating changes in Poisson's ratio and elastic modulus seems to have a significant influence on the slope safety evaluation.