测绘学报 ›› 2022, Vol. 51 ›› Issue (10): 2160-2170.doi: 10.11947/j.AGCS.2022.20220297

• • 上一篇    下一篇

基于深度学习的滑坡位移时空预测

罗袆沅1, 蒋亚楠1,2, 许强1, 廖露3, 燕翱翔2, 刘陈伟2   

  1. 1. 成都理工大学地质灾害防治与地质环境保护国家重点实验室,四川 成都 610059;
    2. 成都理工大学地球科学学院,四川 成都 610059;
    3. 四川测绘地理信息局测绘技术服务中心,四川 成都 610081
  • 收稿日期:2022-05-05 修回日期:2022-09-08 发布日期:2022-11-05
  • 通讯作者: 蒋亚楠 E-mail:jiangyanan@cdut.edu.cn
  • 作者简介:罗袆沅(1997—),男,博士生,主要从事地质灾害评价与预测研究。E-mail:luohiuyuancdut@163.com
  • 基金资助:
    国家自然科学基金(41877254; 41977255);第二次青藏高原综合科学考察研究(2019QZKK0201);国家重点实验室开放基金(SKLGP2017K016)

A spatio-temporal network for landslide displacement prediction based on deep learning

LUO Huiyuan1, JIANG Yanan1,2, XU Qiang1, LIAO Lu3, YAN Aoxiang2, LIU Chenwei2   

  1. 1. State Key Laboratory of Geological Hazard Prevention and Geological Environment Protection, Chengdu University of Technology, Chengdu 610059, China;
    2. Institute of Earth Sciences, Chengdu University of Technology, Chengdu 610059, China;
    3. Surveying and Mapping Technology Service Center of Sichuan Bureau of Surveying, Mapping and Geoinformation, Chengdu 610081, China
  • Received:2022-05-05 Revised:2022-09-08 Published:2022-11-05
  • Supported by:
    The National Natural Science Foundation of China (Nos. 41877254; 41977255); The Second Qinghai Tibet Plateau Comprehensive Scientific Investigation of Research Support (No. 2019QZKK0201); The State Key Laboratory Open Fund (No. SKLGP2017K016)

摘要: 滑坡变形监测数据是认识滑坡变形演化规律的直接依据,对该数据深度挖掘是实现滑坡灾害预警预报的有力保障。现有的滑坡位移预测模型多局限于单个监测点的时序预测,且未考虑监测点间的空间相关性。针对上述问题,本文提出了一种基于深度学习的滑坡位移时空预测模型:首先,构建表达所有点间空间相关性的加权邻接矩阵;其次,引入外界影响因素加强属性特征矩阵,以构建图结构数据;最后,采用集合图卷积网络(GCN)和门控循环单元(GRU)的深度学习模型,并通过多组试验寻找最优超参数,实现滑坡位移的时空预测。本文模型结果的均方根误差为4.429 mm,与对比模型相比至少降低了47.3%。而消融试验结果也显示,引入外界影响因素的属性增强可进一步提高模型的预测性能,均方根误差相对于未属性增强结果减少了28.4%。结果表明,该方法可用于滑坡位移或其他地质灾害中同样具有时空关联属性的观测量的时空预测。

关键词: 滑坡, 图卷积网络, 时序预测, 门控循环单元, 空间相关性

Abstract: Landslide deformation monitoring data is the direct basis for understanding the evolution law of landslide deformation, and the deep mining of this data is a powerful guarantee to realize the early warning and prediction of landslide disaster. The existing landslide prediction models are mostly limited to the time-series displacement prediction of a single monitoring point and do not consider the spatial correlation among monitoring points. To address the above problems, this paper proposes a spatio-temporal prediction model for landslide displacement based on deep learning: Firstly, the weighted adjacency matrix expressing the spatial correlation among all points in the interpretation is constructed; Secondly, the external influences are introduced to strengthen the attribute feature matrix in order to construct the graph structure data; Finally, this model of ensemble graph convolutional network (GCN) and gate recurrent unit (GRU) is used, and the optimal hyper-parameters are found through multiple sets of experiments .Compared with the comparison model, the root mean square error(RMSE) of the proposed model is 4.429 mm, which is at least 47.3% lower. The ablation experiment results also show that the attribute augmentation with the introduction of external influences can further improve the prediction performance of the model, and the RMSE is reduced by 28.4% compared with the results without attribute augmentation. The results suggest that the method can be used for spatio-temporal prediction of landslide displacements or other observed quantities in geological hazards that also have spatio-temporal correlation properties.

Key words: landslide, graph convolution network, time series prediction, gate recurrent unit, spatial correlation

中图分类号: