基于信息量-支持向量机组合模型的东莞市地面沉降灾害易发性评价  

Susceptibility Evaluation of Land Subsidence Disaster in Dongguan City Based on IM-SVM Combined Model

作　　者：方宝 张贤 张丽平 刘岳霖 左昌群[2] FANG Bao;ZHANG Xian;ZHANG Liping;LIU Yuein;ZUO Changqun(Ninth Geological Brigrade of the Guangdong Geological Burerau,Dongguan,Guangdong 523000,China;Faculty of Engineering,China University of Geosciences,Wuhan,Hubei 430074,China)

机构地区：[1]广东省地质局第九地质大队,广东东莞523000 [2]中国地质大学(武汉)工程学院,湖北武汉430074

出　　处：《水利与建筑工程学报》2024年第2期149-157,共9页Journal of Water Resources and Architectural Engineering

基　　金：广东省地勘事业发展基金项目(2019202)。

摘　　要：易发性评价是地面沉降灾害风险识别及减避控制的靶向关键技术问题,为解决单一评价模型存在的过拟合问题,以东莞市地面沉降发育重点区为对象,采用相关系数法筛选并构建由软土层厚度、填土层厚度、与断裂带距离、建筑密度和道路密度组成的评价因子组合,开展基于信息量-支持向量机组合模型的地面沉降易发性评价,并结合地理信息系统(GIS)划分出高(Ⅰ)、中(Ⅱ)和低(Ⅲ)3个易发性等级分区。结果表明:使用受试者工作特征曲线(ROC)和历史灾点验证法校验,信息量-支持向量机(IM-SVM)组合模型的AUC值为0.915,位于高、中易发性分区内的灾点数为90.57%,较单一信息量模型(IM)和支持向量机模型(SVM)具有更高的评价精度。基于数据驱动的机器学习法在对灾害评价分类方面有更强的优越性,进一步选取相同原理、性能差异不大的学习器进行组合更能实现评价可靠性的提高。Susceptibility evaluation is a key technical problem for the identification and mitigation control of land subsidence disaster risk.In order to solve the overfitting problem of a single evaluation model,taking the developed subsidence area of Dongguan City as the object,the correlation coefficient method was adopted to select the evaluation factors which was consisting of the thickness of soft soil layer,the thickness of fill soil layer,the distance from the fracture zone,Injilding density and road density.The land subsidence susceptibility evaluation was carried out based on the Information Model(IM)and Support Sector Machine Model(SVM)combined model.The geographic information system(CIS)was also used to achieve three susceptibility levels of high,medium and low.It is showed that using receiver operating characteristic(ROC)curve and historical disaster point verification,the AUC value of IM-SVM combined model was 0.915,the number of disaster points located in high and medium grade zones was 90.57%,and the combined model has higher prediction accuracy than the single information model(IM)and support vector machine(SVM).The machine learning method based on data-driven has more advantages in disaster evaluation and classification,and the reliability of evaluation can be improved by using the same principle and similar performance.

关 键 词：东莞市 地面沉降 信息量-支持向量机组合模型 机器学习 易发性评价 

分 类 号：X43[环境科学与工程—灾害防治]