东北大学学报（自然科学版）

东北大学学报:自然科学版 ›› 2020, Vol. 41 ›› Issue (5): 609-615.DOI: 10.12068/j.issn.1005-3026.2020.05.001

• 资源与土木工程 •    下一篇

基于深度信念网络的滑坡敏感性评价

王卫东1,2， 何卓磊1， 韩征1， 钱于3   

  1. (1.中南大学 土木工程学院， 湖南 长沙410075； 2.中南大学 重载铁路工程结构教育部重点实验室， 湖南 长沙410075；3.南卡罗来纳大学 土木与环境工程系， 哥伦比亚29208)
  • 收稿日期:2019-08-26 修回日期:2019-08-26 出版日期:2020-05-15 发布日期:2020-05-15
  • 通讯作者: 王卫东
  • 作者简介:王卫东(1971-) ，男，江西上饶人，中南大学教授，博士生导师.
  • 基金资助:
    国家自然科学基金资助项目(51478483，41702310); 国家重点基础研发计划项目(2018YFD1100401).

LandslidesSusceptibilityAssessmentBasedonDeepBeliefNetwork

WANG Wei-dong1,2， HE Zhuo-lei1， HAN Zheng1， QIAN Yu3   

  1. 1.School of Civil Engineering，Central South University， Changsha 410075， China; 2.The Key Laboratory of Engineering Structures of Heavy Haul Railway， Ministry of Education， Central South University， Changsha 410075， China; 3.Department of Civil and Environmental Engineering， The University of South Carolina， Columbia 29208， America.
  • Received:2019-08-26 Revised:2019-08-26 Online:2020-05-15 Published:2020-05-15
  • Contact: HAN Zheng
  • About author:-
  • Supported by:
    -

RichHTML

1

PDF (PC)

893

摘要： 滑坡敏感性评价中各致灾因子之间存在复杂非线性关系，传统的评价模型难以揭示该类复杂关系，以致评价结果精度受限.基于文献调查与实地调研，选取高程、地貌类型、岩性、坡度、与构造线距离、与水系距离和年均降雨量为主要致灾因素，在地理信息系统(GIS)中建立了基于深度信念网络(DBN)模型的区域滑坡敏感性区划模型，并以四川区域为例进行了实例分析.最后通过ROC曲线特征将评价结果与逻辑回归(LR)和人工神经网络(BPNN)模型评价结果进行了对比分析，并探讨了各评价模型对不同致灾因子的响应.研究表明DBN模型具有更高精度以及较低的假阳性率和假阴性率，更适合于大区域、复杂致灾因素的区划滑坡敏感性评价工作.

关键词: 地理信息系统, 滑坡敏感性评价, 深度学习, 深度信念网络, ROC曲线

Abstract: Complex non-linear relationships exist among causing factors in landslides susceptibility assessment. Traditional assessment models is difficult to reveal such complex relationships， and limit the accuracy of assessment results. Based on literature review and field survey， the altitude， landform， lithology， slope， distance to tectonic line， distance to drainage network and annual average rainfall were chosen as the main causing factors. A regional landslides susceptibility mapping model based on deep belief network (DBN) model in geographic information system (GIS) was established， and Sichuan Province was taken as an example. Finally， through the characteristics of ROC curves， the assessment results were compared with logistic regression (LR) and BP neural network (BPNN)， and the response of each model to different causing factors was discussed. The results show that the DBN model has high accuracy， low false positive rate and false negative rate， and is suitable for landslides susceptibility assessment in large area with complex causing factors.

Key words: geographic information system(GIS), landslides susceptibility assessment, deep learning, deep belief network(DBN), ROC curve

中图分类号: