双向门控循环单元网络的滑坡易发性评价  

The susceptibility assessment of landslide based on Bi-GRU network

作　　者：高轩宇 王继周 毛曦[1,2] 赵占骜 路文娟[2] GAO Xuanyu;WANG Jizhou;MAO Xi;ZHAO Zhanao;LU Wenjuan(College of Geodesy and Geomatics,Shandong University of Science and Technology,Qingdao,Shandong 266590,China;Chinese Academy of Surveying&Mapping,Beijing 100036,China;Faculty of Geomatics,Lanzhou Jiaotong University,Lanzhou 730070,China)

机构地区：[1]山东科技大学测绘与空间信息学院,山东青岛266590 [2]中国测绘科学研究院,北京100036 [3]兰州交通大学测绘与地理信息学院,兰州730070

出　　处：《测绘科学》2023年第4期221-230,共10页Science of Surveying and Mapping

基　　金：国家重点研发计划项目(SQ2020YFF0426312);国家重大测绘工程“全球地理信息资源建设与维护更新”项目(AR1936)。

摘　　要：针对传统滑坡易发性评价模型在训练中直接将影响因子叠加输入,未考虑特征重要程度以及在学习过程中前后因子间的相互作用问题,该文提出一种新型滑坡易发性评价方法。以色东普沟为研究对象,首先采用多重共线性分析筛选影响因子并通过XGBoost算法计算影响因子特征重要度以确定因子输入模型的顺序。然后构建了顾及因子序列输入后前后相互作用的双向门控循环单元(Bi-GRU)的滑坡易发性评估模型,提升预测结果的准确性。最后将结果与门控循环单元(GRU)、支持向量机(SVM)和多层感知器(MLP)3种模型从深度学习常用的评价指标以及滑坡易发性评价结果两方面进行对比。实验结果表明,该文方法的准确性与可靠性高于传统模型,可以较为准确地预测未来发生滑坡范围。In view of the fact that the traditional landslide susceptibility mapping model directly superimposes the influencing factors in the training,without considering the importance of the features and the interaction between the factors before and after the learning process,a new landslide susceptibility mapping method was proposed in this paper.Taking Sedongpu gully as the research object,firstly,the multiple collinear analysis was used to screen the influencing factors and the XGBoost algorithm was used to calculate the characteristic importance of the factors to determine the order of the factor input model,and then the landslide susceptibility mapping model based on Bi-GRU was constructed.The model taken into account the interaction of the factors before and after the input of the factor sequence,and improved the accuracy of the prediction results.Finally,from the two aspects of the evaluation indicators commonly used in deep learning and the evaluation results of landslide susceptibility maps,the results were compared with the results of GRU,SVM and MLP models.The results showed that the accuracy and reliability of the proposed method was higher than that of the traditional model,which could accurately predict the range of landslides in the future.

关 键 词：滑坡易发性 深度学习 双向GRU网络 特征选择 

分 类 号：P642.22[天文地球—工程地质学]