节点文献

基于机器学习混合模型的滑坡易发性评价

Landslide Susceptibility Assessment Based on Hybrid Model of Machine Learning

  • 推荐 CAJ下载
  • PDF下载
  • 不支持迅雷等下载工具,请取消加速工具后下载。

【作者】 邓念东李宇新崔阳阳石辉郭亚雷

【Author】 DENG Nian-dong;LI Yu-xin;CUI Yang-yang;SHI Hui;GUO Ya-lei;School of Geology and Environment, Xi’an University of Science and Technology;

【机构】 西安科技大学地质与环境学院

【摘要】 安康市汉滨区地质环境脆弱,滑坡频发对当地居民生命财产安全造成严重威胁,针对该区域进行滑坡易发性评价是滑坡防治的有效措施。自适应提升模型和随机森林模型作为新颖的集成学习方法被应用至中外滑坡易发性评价研究中,但基于两者的混合模型在滑坡易发性中的应用研究尚未开展。为对比混合模型与单一模型的滑坡易发性评价精度,根据地质灾害详查资料圈定509处滑坡,结合研究区地质环境背景,选取高程、坡度、坡向、年均降雨量、地层岩性等13类因子进行评价。受试者工作特性曲线(receiver operating characteristic curve, ROC)结果表明,同单一模型相比,混合模型的训练集正确率和验证集预测率均为最高;混合模型的高易发区滑坡密度达到1.94,高于随机森林(1.86)和自适应提升模型(1.68);通过区内三处历史滑坡进行验证,结果显示区划结果与滑坡分布相吻合,说明自适应提升-随机森林混合模型可作为滑坡易发性评价的新方法,其区划结果可为滑坡防治与土地利用规划提供借鉴。

【Abstract】 Hanbin district of Ankang city, the geological environment is fragile. The lives and property of local residents have been seriously threatened by frequent landslides. It is an effective measure to evaluate the susceptibility of landslides in this area. As a novel integrated learning method, adaptive boosting model and random forest model have been applied to the evaluation of landslide susceptibility at home and abroad. However, the application of the hybrid model based on them has not been carried out yet. In order to compare the accuracy of landslide susceptibility evaluation between the hybrid model and the single model, 509 landslides were selected based on the detailed survey data of geological hazards. Combined with the geological environment background of the study area, 13 types of factors such as elevation, slope angle, slope aspect, average annual rainfall and lithology were selected for evaluation. The results of receiver operating characteristic curve showed that the hybrid model has the highest accuracy of training set and prediction rate of validation set compared with the single model. The landslide density of the hybrid model reached 1.94, which was higher than that of the random forest(1.86) and the adaptive boosting model(1.68). Three historical landslides in the area were verified and the results showed that the zoning results were consistent with the landslide distribution, which indicated that the adaptive boosting-random forest hybrid model could be used as a new method for landslide susceptibility evaluation and the zoning results could provide reference for landslide prevention and land use planning.

【基金】 国家自然科学基金(41602359);青海省青藏高原北部地质过程与矿产资源重点实验室基金(2019-KZ-01)
  • 【文献出处】 科学技术与工程 ,Science Technology and Engineering , 编辑部邮箱 ,2022年14期
  • 【分类号】TP181;P642.22
  • 【下载频次】437
节点文献中: 

本文链接的文献网络图示:

本文的引文网络