基于逻辑回归和支持向量机耦合模型的滑坡易发性分析  

Landslide susceptibility analysis based on the coupling model of logistic regression and support vector machine

作　　者：李成林 刘严松[1,3,4] 赖思翰 王地 何星慧 刘琦 LI Chenglin;LIU Yansong;LAI Sihan;WANG Di;HE Xinghui;LIU Qi(State Key Laboratory of Geohazard Prevention and Geoenvironment Protection(Chengdu University of Technology),Chengdu 610059,China;Cecep Construction Engineering Design Institute Limited,Chengdu 610059,China;Chengdu Geological Survey Center,China Geological Survey,Chengdu 610081,China;China University of Geosciences(Beijing),Beijing 100083,China;Sichuan Sumhope Spatial Technology Co.,Ltd.,Chengdu 610094,China)

机构地区：[1]地质灾害防治与地质环境保护国家重点实验室(成都理工大学),四川成都610059 [2]中节能建设工程设计院有限公司,四川成都610059 [3]中国地质调查局成都地质调查中心,四川成都610081 [4]中国地质大学(北京),北京100083 [5]四川三合空间科技有限公司,四川成都610094

出　　处：《自然灾害学报》2024年第2期75-86,共12页Journal of Natural Disasters

基　　金：四川省教育厅基金项目(18ZB0065);四川省自然资源厅基金项目(KJ2016-16);国家自然科学基金项目(41402159);中国地质调查局地调项目(DD20221697)。

摘　　要：滑坡灾害的发生具有累进性,进行滑坡易发性评价是防灾减灾的前提。以四川省旺苍县为例,使用频率比法判断12个滑坡影响因子的各分级区间滑坡敏感性,经波段集统计确定11个滑坡影响因子作为滑坡易发性评价因子,通过建立逻辑回归-支持向量机(logistic regression-support vector machine,LR-SVM)耦合模型,搭建滑坡易发性评价体系,完成旺苍县滑坡易发性评价并进行模型精度比较。研究结果表明:逻辑回归-支持向量机耦合模型的评价指标结果均优于逻辑回归模型,易发性分区结果更合理,预测精度更高;在低易发区选取非滑坡点为提高滑坡易发性评价性能作用明显;研究区内道路、高程和NDVI对滑坡发育的敏感性较强;高易发区主要分布于低海拔的水系和道路两侧。The occurrence of landslide disasters is progressive,and evaluating the susceptibility of landslides is a prerequisite for disaster prevention and reduction.The Wangcang County,Sichuan Province is taken as an example and the frequency ratio method is used to determine the landslide sensitivity of 12 landslide impact factors in each grading interval.Through the band set statistics,11 landslide impact factors are determined as landslide susceptibility evaluation factors.By establishing a logistic regression support vector machine(LR-SVM)coupling model,a landslide susceptibility evaluation system is established,completing the evaluation of landslide susceptibility in Wangcang County and comparing model accuracy.The results show that the evaluation index results of the LR-SVM coupling model are better than those of the logistic regression model,and the susceptibility zoning results are more reasonable and the prediction accuracy is higher.Selecting non-landslide points in low risk areas has a significant effect on improving the evaluation performance of landslide susceptibility.The sensitivity of roads,elevations,and NDVI in the study area to landslide development is strong.High risk areas are mainly distributed in low altitude water systems and on both sides of roads.

关 键 词：滑坡易发性评价 逻辑回归 支持向量机 耦合模型 旺苍县 

分 类 号：P642.22[天文地球—工程地质学] X43[天文地球—地质矿产勘探]