Landslide Susceptibility Evaluation Based On Optimized Support Vector Machine Model

In recent years,human engineering activities are increasingly frequent,which can not avoid the impact on the natural environment,which leads to the occurrence of landslide disasters,which restricts the sustainable development of economy and society.Therefore,it is of great practical significance to evaluate the landslide susceptibility.In the evaluation of vulnerability,the sample data set is the basis of the vulnerability model,but the research on the selection of sample data sets is relatively insufficient.For the model itself,the ability of fitting and generalization of the model is determined by the super parameters,but the selection of the super parameters in the current model is lack of science.In view of the above problems,this paper,relying on the national key R&D plan"integrated integrated intelligent service research and application demonstration for disaster reduction",takes Shangluo city of Shanxi Province as the research area,and conducts the following research:(1)The paper proposes a factor screening method which takes into account the correlation and weight of factors,and removes redundant factors in sample set,and improves the efficiency of the model;The method of generating"non landslide point"under multi constraints is constructed to avoid the wrong selection of"non landslide point",which lays the sample data foundation for the subsequent construction of the vulnerability model.(2)According to the characteristics of landslide sample data set,based on support vector machine(SVM),the evaluation model of landslide vulnerability is constructed;The paper introduces the artificial bee colony algorithm(ABC)to optimize the error rate of landslide sample classification as the objective function by SVM vulnerability model.When the error rate of landslide sample classification reaches the lowest,the global optimal penalty factor and kernel width are found.The optimization results are replaced into SVM model to build the vulnerability model.(3)The model accuracy is evaluated from three aspects:the probability zoning map,the density statistics of landslide points,the receiver operator characteristic(ROC)and so on.The results show that the optimized model needle is more accurate in identifying the high-risk area,and the identification of the high-risk area is more accurate Compared with the classical model,the density of landslide points in the high prone area increased to 0.12/km~2 and 0.07/km~2respectively.The distribution of landslide points is more consistent with the actual situation.The ROC accuracy of the two models is 0.925 and 0.896 respectively,and the optimized model precision is significantly higher than that of the classical model.(4)The results of the evaluation of the vulnerability of Shangluo City show that the high incidence area of landslide is distributed in the southeast and extends to the southwest,and sporadic in the north;From the view of topography,landslides are mostly distributed along the valley in Qinling mountain area.The paper has 30 diagrams,8 tables and 52 references.