Research On Stability Evaluation Method Of Soil Slope Based On Improved Support Vector Machine

With the rapid development of Chinese economy and the increasing number of large-scale infrastructure projects,the problem of slope stability has become one of the research hotspots in geotechnical engineering.The instability of the slope causes a large number of casualties and property losses every year.It is of great significance to evaluate the stability of the slope accurately,efficiently and reliably.The slope is a complex open system,and the traditional deterministic analysis method has some limitations.In recent years,the application of intelligent method in the field of geotechnical engineering has brought new ideas to the analysis for slope stability analysis.Among them,support vector machines(SVM)have unique advantages in dealing with small samples,high dimensions and non-linear problems,which has become a hot research topic in machine learning algorithms.Through the study of the slope samples,combined with the influence factors of slope stability,the introduction of the intelligent optimization algorithm,modern statistics theory and nonlinear science theory.The evaluation model of soil slope stability is established to analyze and evaluate slope stability scientifically.This article takes the soil slope as the research object,adopts the method of support vector machine to study,establishes the soil slope stability evaluation model based on the influencing factors,and improves the model parameter selection.On account of the improved support vector machine model,the nonlinear relationship between displacement information and safety factor is established.The displacement monitoring information evaluates the slope stability in real time.The specific research contents are as follow:Firstly,analyze and selecte the factors affecting the stability of soil slope.By using the attribute reduction theory and attribute importance theory in rough set theory(RS)to avoid too much subjective elements,the main controlling factors affecting the stability of soil slope are selected and analyzed by single factor and multi-factor sensitivity.Secondly,combined with the selected influencing factors,the support vector classification machine(SVC)and the support vector regression machine(SVR)are established to predict the slope stability,comprehensive evaluation of slopes using two indicators of steady state and safety factor.Considering that the parameter selection of SVM will have a great influence on the modeling results,genetic algorithm,particle swarm optimization algorithm and grid search algorithm are respectively adopted to optimize the selection of kernel function parameter g and penalty factor C of SVM(including SVC and SVR)to improve the prediction accuracy.On the basis of comparing the prediction results of the three optimization methods,a grid particle optimization support vector machine model was established to further improve the prediction accuracy and efficiency,and it was applied to the stability evaluation of xibang soil slope in Sijiaying Iron Mine.The prediction results are consistent with the calculation results of the limit equilibrium method and the strength reduction method.Finally,the mesh particle optimization support vector machine is used to establish the stability evaluation model of soil slope based on displacement information.The real-time shear strength parameters are obtained by back analysis of the displacement monitoring information,and the corresponding safety factor is calculated by the strength reduction method,and then the dynamic safety factor of the soil slope is obtained.The improved support vector machine model is used to establish the nonlinear relationship between displacement value and slope safety factor to realize the real-time evaluation of soil slope stability.This diseertations has a certain theoretical significance and practical value for improving the evaluation accuracy of the stability of soil slope with support vector machine and popularizing the intelligence degree of slope stability analysis.