Slope Reliability Analysis Based On SVM

Slope reliability analysis relates to a wide range of engineering fields, is an important problem in Geotechnical Engineering. Being unable to consider various uncertainties existed in slope, the result of traditional Safety Factor Method can not reflect the risk of slope accurately. The research presented in this thesis combines support vector machine (SVM) with stability analysis. The emphases are as follows:(1).Establish a forecast model of slope safety factor to solve the problem that the safety factor of some complex slope whose performance function can not be expressed explicitly has to be calculated via softwares. The proposed method fits the actual performance function of slope via SVM, based on the training sample,which is formed by performing deterministic computations at some designed points.The result shows that SVM can be used to predict the safety factor of slope accurately.(2).SVM, first-order reliability method (FORM), second-order reliability method (SORM), Monte Carlo simulation (MCS) are applied to slope reliability analysis. Based on the reliability index and the design point obtained from FORM,the SVM are optimized via an iterative algorithm.Based on the optimized SVM,the probability of failure of slope can be calculated using SORM and MCS.By applying this method to two typical slopes and comparing the results with other reliability methods, we can conclude that, the proposed method are both accurate and efficient.(3).The DP point in U-space got from FORM can reflect the parametric sensitivity,and the principal curvature got from SORM can reflect the nonlinearity of the actual performance function.(4).The influence of the distribution pattern of parameters,the correlation between parameters,and the variability of parameters on the result of reliability analysis are studied via a typical slope.Distribution pattern has nearly no influence on the result when different distribution functions have similar PDF. Slope reliability index decreases with the increase of the variation of parameters.The impact of the correlation between parameters on the reliability index depends on the characteristics of the correlation,the limit state surface as well as the parameter sensitivity.(5). Random field is introduced into slope reliability analysis and implemented by coupling interpolation method with SVM. In this way, the spatial variability of soil can be included in the reliability analysis.