Reliability Algorithm Of Slope Stability Based On Metamodel

In the analysis on the stability reliability of the slope engineering, the function ofthe slope performance deduced with the limit equilibrium method tends to be aimplicit function, which has no obvious expression and can not be solved by thetraditional reliability-analysis method, such as the first-order two-moment method.And the method of Monte Carlo, widely used in solving the above mentioned problemis difficult to widely spread in the practical engineering due to its enormouscomputation and low computational efficiency. To solve the problem, the implicitfunction can be made explicit by using metamodels before obtaining the reliabilityindex with traditional methods. In this paper, the main study is focus on theapplication of the Kriging metamodel from the Geostatistics on the reliability-analysisof the performance function. And the study is conducted with two different ways.Firstly, trough combining the Latin hypercube design, Kriging metamodel andthe method of JC, a new calculation method for the implicit function of the slopeperformance is proposed and the relevant technical details and operation steps are alsogiven. The optimum sampling frequency of the Latin hypercube design is found basedon the analysis on a slope case. The comparative analysis between the presentedmethod and the method of Monte Carlo indicates that the new method has a clearercalculating process and less computation, and its computational accuracy meet therequirement of the engineering and has comparatively strong practicality inengineering.Secondly, aiming at the interdependency of the Kriging metamodel on the designof experiment, a policy of initiative learning is proposed by introducing the ActiveLearning System. The new policy can search the best training samples actively, bywhich the he Kriging metamodel on the function of the slope performance with higheraccuracy can be established. Then, through combining the performance function andthe method of Monte Carlo, the computational effort is reduced greatly relative to thecase of only Monte Carlo. The case of slope mentioned above is used to verify thefeasibility and accuracy of the new method as well.Lastly, the two methods proposed in this paper are used to analyze thestability-reliability of a cutting slope of the Henggui Highway, respectively. Theresults calculated from the two methods are basically consistent, which verify the feasibility and the rationality of the methods proposed in this paper, furtherly.