基于支持向量机的滑坡易发性评价  

Landslide susceptibility mapping based on support vector machine models

作　　者：王倩 薛云 张维 龙岳红 周松林[3] WANG Qian;XUE Yun;ZHANG Wei;LONG Yuehong;ZHOU Songlin(School of Municipal and Surveying Engineering,Hunan City University,Yiyang,Hunan 413000,China;School of Traffic and Transportation Engineering,Changsha University of Science&Technology,Changsha,Hunan 410114,China;Hunan City University Design and Research Institute CO.,Ltd.,Yiyang,Hunan 413000,China)

机构地区：[1]湖南城市学院市政与测绘工程学院,湖南益阳413000 [2]长沙理工大学交通运输工程学院,长沙410114 [3]湖南城市学院设计研究院有限公司,湖南益阳413000

出　　处：《湖南城市学院学报（自然科学版）》2021年第1期22-28,共7页Journal of Hunan City University:Natural Science

基　　金：国家自然科学基金项目(51678226);湖南省科技厅项目(2019TP2079,2018TP1042,2017TP2006);湖南省教育厅科研项目(HNKCSZ-2020-0495);湖南省住建厅科研项目(2016-42)。

摘　　要：应用支持向量机算法对湖南省靖州县的滑坡易发性进行评价.首先,通过实地调查、卫片判译及滑坡历史记录,共发现滑坡102处及非滑坡点100处,随机用70%数据来训练模型,30%数据来验证模型;其次,选取坡度、坡向、高度、河流距离、断层距离、公路距离、土地利用和人类活动强度8个地质灾害影响因子作为地质灾害易发性评价指标;然后,用粒子群算法优化的参数和4种核函数分类器训练分类模型,利用训练好的4个模型计算滑坡易发性指数,使用自然断点法将易发性指数值划分为高、中、低3个易发等级;最后,利用ROC曲线,并以信息量法作为比较算法,验证滑坡易发性图的合理性.结果表明:4种支持向量机模型的成功率分别为83.12%(RBF-SVM)、82.61%(PL-SVM)、81.53%(LN-SVM)和79.98%(SIG-SVM),预测率分别为77.87%(RBF-SVM)、77.34%(PL-SVM)、76.89%(LN-SVM)和76.01%(SIG-SVM);信息量法的成功率和预测率分别为78.56%和75.76%.In this study,a support vector machine algorithm is used to evaluate the landslide susceptibility in Jingzhou County,Hunan Province.Firstly,through field surveys,satellite photo interpretation and historical records of landslide,102 landslides,and 100 non-slide points are found.70%data are randomly used to train the model and 30%data are used to verify the model.Secondly,8 factors including slope,aspect,height,river distance,fault distance,road distance,land use,and human activity intensity are selected as the evaluation index of susceptibility.Then,the SVM model is trained with the parameters optimized by the ACO algorithm and four different kernel functions.The trained four models were used to calculate the landslide susceptibility index,and the natural breakpoint method was used to divide the susceptibility index values into high,medium,and low levels.Finally,the ROC curve and the comparison algorithm of the information method are used to verify the rationality of landslide susceptibility mapping.The results show that the success rates of the four SVM models are 83.12%(RBF-SVM),82.61%(PL-SVM),81.53%(LN-SVM),and 79.98%(SIG-SVM),and the prediction rates are 77.87%(RBF-SVM),77.34%(PL-SVM),76.89%(LN-SVM),and 76.01%(SIG-SVM).The success rate and prediction rate of the information method are 78.56%and 75.76%respectively.

关 键 词：滑坡 易发性评价 支持向量机 靖州 

分 类 号：TU413.6[建筑科学—岩土工程]