基于机器学习模型的斜坡地质灾害易发性评价  

SUSCEPTIBILITY EVALUATION OF SLOPE GEOLOGICAL HAZARD BASED ON MACHINE LEARNING MODELS

作　　者：周修波 李永红 陈建平 何意平 姬怡微 蒙晓 张辉 ZHOU Xiu-bo;LI Yong-hong;CHEN Jian-ping;HE Yi-ping;JI Yi-wei;MENG Xiao;ZHANG Hui(Shanxi Institute of Geo-Environment Monitoring,Xi'an 710054,China)

机构地区：[1]陕西省地质环境监测总站,陕西西安710054

出　　处：《甘肃地质》2024年第4期63-75,共13页Gansu Geology

基　　金：陕西省自然资源厅2021年度地质灾害综合防治体系建设项目(陕自然资勘发[2021]42号)。

摘　　要：开展城镇地质灾害风险调查评价,亟需选择适宜精细化城镇地质灾害风险评价的影响因子和评估模型,提高评价精度,以适应地质灾害“隐患点+风险区”双控管理需求。在汉阴县漩涡镇镇域调查评价尺度下,从斜坡孕灾地质环境条件调查中提取的地形地貌、地质构造、水、工程地质岩组、斜坡结构、人类工程活动等6大类16项孕灾因子中选取10项关键因子参与易发性建模。以斜坡为评价单元,采用确定性系数模型(CF)联接到逻辑回归模型(LR)、支持向量机模型(SVM)和随机森林模型(RF)构建耦合CF-LR、CF-SVM和CF-RF模型,评价地质灾害易发性。结果表明,3种模型AUC值分别为0.879、0.868、0.927,Kappa系数分别为0.622、0.609、0.630,均具有良好的空间分异能力和评价精度,CF-RF模型评估能力最优。研究结果对探索城镇地质灾害风险评价体系,构建机器学习评价方法具有一定实践指导意义。In recent years,Shaanxi Province has deployed a new round of urban geological disaster risk investigation and evaluation.It is urgent to select the impact factors and vulnerability assessment models suitable for refined urban geological disaster risk evaluation to improve the evaluation accuracy to meet the dual control management needs of"hidden danger points+risk areas"of geological disasters.Under the scale of investigation and evaluation in Xuanwo Town,Hanyin County,10 key factors were selected from 16 factors in 6 categories,including topography,geological structure,water,engineering geological rock group,slope structure,and human engineering activities,extracted from the survey of slope pregnancy and environmental conditions to participate in susceptibility modeling.With the slope as the evaluation unit,the Certainty Factor(CF)is connected to the Logistic Regression(LR)、the Support Vector Machine(SVM)and the Random Forest(RF)to build a coupled CF-LR,CF-SVM and CF-RF model to evaluate the susceptibility of geological disasters.The results show that the AUC values of the three models are 0.879,0.868 and 0.927 respectively,and the Kappa coefficients are0.622,0.609 and 0.630 respectively,all of which have good spatial separation ability and evaluation accuracy,and the evaluation ability of the CF-RF model is optimal.The results of the research have certain practical guiding significance for exploring the urban geological disaster risk evaluation system and building machine learning evaluation methods.

关 键 词：地质灾害 易发性评价 斜坡单元 机器学习 

分 类 号：P642.22[天文地球—工程地质学]