基于CF-BPNN耦合模型的益湛铁路沿线滑坡危险性评价  

Risk assessment of landslides along the Yizhan railway based on the CF-BPNN coupling model

作　　者：唐学武 刘耕 邵磊 姚灯 陈东旭 田优平 Tang Xuewu;Liu Geng;Shao Lei;Yao Deng;Chen Dongxu;Tian Youping(Hunan Earthquake Agency,Hunan Provincial Earthquake Risk Prevention and Control Center,Changsha 410000)

机构地区：[1]湖南省地震局,湖南省震灾风险防治中心,长沙410000

出　　处：《地质科学》2024年第5期1470-1486,共17页Chinese Journal of Geology(Scientia Geologica Sinica)

基　　金：湖南省地震局防震减灾科研课题项目(编号:202002)资助。

摘　　要：滑坡是一种比较常见的地质灾害,极易造成严重的人员伤亡和财产损失,特别是铁路沿线发育的一系列滑坡给生命线工程带来了极大的风险隐患。本文以益湛铁路益阳至娄底段沿线为研究对象,基于第一次自然灾害风险普查的地质灾害结果和野外地质调查数据,从地形地貌、区域地质、水文地质、人类活动等4个方面,提取16类地质环境因子构建滑坡危险性评价体系,引入确定性系数模型(CF)对传统的BP神经网络模型(BPNN)进行改进,开展滑坡危险性评价,以增强BPNN模型的性能,提高预测的准确率。在此基础上总结研究区的滑坡分布规律特征,其中高程、岩性、道路对研究区的滑坡分布具有重要影响。最后通过ROC曲线将改进的耦合模型与单一的CF模型和BPNN模型进行对比分析。结果表明,CF-BPNN模型的AUC值为0.849,CF模型的AUC值为0.754,BPNN模型的AUC值为0.837,这表明改进的耦合模型较单一模型效果更佳,预测结果准确率更高。研究结果可为近场生命线工程的滑坡风险分析提供信息支撑。Landslide is a common geological hazard that can easily cause serious casualties and property damage,especially the series of landslides developed along the railway line,which pose great risks and hidden dangers to lifeline engineering.This article takes the Yiyang to Loudi section of the Yizhan railway as the research object.Based on the geological disaster results of the first natural disaster risk survey and field geological survey data,16 types of geological environmental factors are extracted from 4 aspects:topography,regional geology,hydrogeology,and human activities to construct a landslide risk assessment system.The deterministic coefficient model(CF)is introduced to improve the traditional BP neural network model(BPNN),and landslide risk assessment is carried out to enhance the performance of the BPNN model and improve the accuracy of prediction.On this basis,summarize the distribution characteristics of landslides in the study area,among which elevation,lithology and roads have important impacts on the distribution of landslides in the study area.Finally,the improved coupling model was compared and analyzed with a single CF model and BPNN model through ROC curves.The results show that the AUC value of the CF-BPNN model is 0.849,the AUC value of the CF model is 0.754,and the AUC value of the BPNN model is 0.837,which indicates that the improved coupling model is better than the single model,and the prediction accuracy is higher.The research results can provide information support for landslide risk analysis of near-field lifeline engineering.

关 键 词：滑坡 确定性系数 BP神经网络 危险性 

分 类 号：P642.22[天文地球—工程地质学]