Non-probability Reliability Analysis Of Soil Slope Stability Based On Ellipsoid Model

Slope reliability analysis has been widely used in geotechnical engineering.Traditional probability reliability has been fully developed.However,in actual geotechnical engineering,there is little statistical information,and it is difficult to obtain the accurate distribution of various parameters,which makes it difficult to apply the probabilistic reliability method.Therefore,it is very necessary to develop a non-probabilistic reliability method to analyze the reliability of slope stability.In general,the non-probabilistic reliability analysis models only need the variation range of random variables.Therefor,the reliability analysis of slopes is not limited to conventional probability analysis method.This paper takes the ellipsoid model as the research object and applies it to the reliability analysis of infinite slope and two-dimensional slope.A reliability analysis of a slope based on extreme learning machine(ELM)has been developed as the implicit function of the slope.The main research contents and results are as follows:1)The characteristics of the ellipsoidal model and the interval model and their differences and connections between them are represented in detail.Based on the Khachiyan method to extract parameter features,the ellipsoid model is determined according to the parameter correlation.Soil test data are used to verify the efficiency and applicability of the method.2)The first order approximation method based on the ellipsoid model is applied to examine the reliability of the one-dimensional infinite slope.Taking the residual soil test data in the Jurong area of Singapore as an example,the difference between the non-probability reliability and the probability reliability is examined,and the interval model and the ellipsoid model are compared.The study shows when there is few test data,the probabilistic reliability methods are subjective in the selection of probability density function and different probability density functions have a significant effect on the calculatied results.The non-probabilistic reliability method can offer a more objective evaluation;The interval model is inconvenient to consider the correlation of parameters.3)For the highly nonlinear limit state function,the first order approximate method has a large error.Thus a second order approximation method is proposed.This method considers the taylor series second-order expansion of limit state function at the design point.Using the average curvature at the design point instead of the original curvature to construct a new failure surface.Based on slope stability analysis method and response surface model,a non-probabilistic reliability method of slopes is developed.This method is applied to the reliability evaluation of two-dimensional slope,and the influence of parameters on the non-probability reliability is discussed.4)In view of the inconvenience of the polynomial response surface model to deal with highly nonlinear problems.A particle swarm optimization(PSO)optimizing ELM model is proposed to construct the response surface function of slope.Based on the ELM model,a non-probabilistic reliability method is developed.This method is applied to the reliability analysis of Yanan loess slope and Heifangtai loess slope.