Geotechnical Classification And Slope Reliability Analysis Methods Based On Support Vector Machines

The small sample, nonlinear and implicit function problems were often encountered in geotechnical classification and stability reliability analysis. For example, the safety coefficient expression of slope stability analysis resulted from limit equilibrium theory is a highly nonlinear implicit form, which result in that the reliability can not be calculated by accurate analytic method. At the present time, the problem was dealt with by response surface method, artificial neural network method and implicit difference method, but there are different limitations in these methods. Support vector machines (SVM) could resolve the practical problems very well, such as small samples, nonlinear, high dimensions and local optimal point issues. Introducing it to geotechnical classification and slope reliability analysis can overcome the deficiencies of above methods effectively.After a detailed discussion on the principle of support vector machine, it was introduced into geotechnical classification and slope reliability analysis. The main study results are showed as follows. First, a nonlinear SVM sand liquefaction model was built based on analyzing the influence factors deeply. Inputing the normalized training sample to the model, and the parameters were revised by "grid-research" method according to the kernel function principle. Then the seriousness of sand liquefaction was predicted, and the result was compared with the one from BP neural network, which demonstrate that the SVM model has the merits of high precision, simple structure, high speed and stable performance. Second, after analyzing and filtering the influence factors of surrounding rock stability, a Polytomies SVM model was established, and the practicability and validity of the model were verified through the example analyzed. Third, a microcomputerized program for stability coefficient calculation based on limit equilibrium theory was developed using VC++6.0 language. Fourth, taking Spnecer and Janbu mode for examples, and based on the fitting ability of SVM, a proximate explicit method of nonlinear stability coefficient under slope limit equilibrium method was studied. According to the explicit expression, the limit state equation was established and the reliability index was solved by JC method. Three instances proceeded, after contrasting the results to the one of Monte-Carlo method, the high accuracy of the approach available was proved.The SVM model has many merits, such as simple mechanism, fast speed, high precision and can be used for small samples and intricate problems of the geotechnical classification and slope reliability analysis, where it is an advanced analysis method, and has important theoretical and practical significance.