基于Stacking集成学习的泥石流易发性评价:以四川省雅江县为例  

Debris flow susceptibility evaluation based on Stacking ensemble learning:a case study in Yajiang,Sichuan Province

作　　者：苏刚 秦胜伍[1] 乔双双[1] 扈秀宇 陈阳 车文超 SU Gang;QIN Sheng-wu;QIAO Shuang-shuang;HU Xiu-yu;CHEN Yang;CHE Wen-chao(College of Construction Engineering,Jilin University,Changchun 130026,China)

机构地区：[1]吉林大学建设工程学院,长春130000

出　　处：《世界地质》2021年第1期175-184,共10页World Geology

基　　金：国家自然科学基金项目(41977221);吉林省科技发展计划项目(20190303103SF)资助。

摘　　要：为给四川省雅江县提供直观准确的泥石流易发性图,将四川省雅江县作为研究区,选用高程、坡度、坡向、地形起伏度、平面曲率、剖面曲率、年平均降雨、到河流的距离、到道路的距离、地形湿度指数、归一化差分植被指数及土壤类型12个评价因子,利用Stacking集成学习框架,结合支持向量机、神经网络和随机森林,建立了一种多模型融合的泥石流预测模型。通过ROC曲线验证了模型的准确性,得出Stacking融合模型、随机森林、神经网络和支持向量机模型的成功率分别是98.1%、96.1%、94.5%、93.4%,预测率分别是95.5%、91.6%、90.6%、89.7%。结果表明:Stacking融合模型精度最高,最适合用于雅江县泥石流易发性评价。In order to provide an intuitive and accurate debris flow susceptibility map of Yajiang in Sichuan Province,Yajiang is taken as the study area and 12 evaluation factors including elevation,slope,slope direction,topographic relief,plan curvature,profile curvature,average annual rainfall,distance to rivers,distance to roads,normalized difference vegetation index,topographic wetness index,and soil type are selected.A multi-model fusion debris flow prediction model is established by using the Stacking ensemble learning framework combined with support vector machine,neural network and random forest.The accuracy of the model was verified by ROC curve.The success rates of the Stacking fusion model,random forest,neural network,and support vector machine model are 98.1%,96.1%,94.5%and 93.4%,and the prediction rates are 95.5%,91.6%,90.6%and 89.7%,respectively.The results show that the Stacking fusion model has the highest accuracy and is most suitable for the evaluation of debris flow susceptibility in Yajiang.

关 键 词：泥石流易发性 Stacking集成学习 随机森林 支持向量机 神经网络 雅江县 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] P642.23[自动化与计算机技术—控制科学与工程]