[1]周 莹a,樊晓一a,b*,等.基于梯度下降算法优化滑坡运动距离预测模型[J].山地学报,2026,(1):74-88.[doi:10.16089/j.cnki.1008-2786.000948]
　ZHOU Yinga,FAN Xiaoyia,b*,et al.Optimization of Landslide Runout Prediction Model Based on Gradient Descent Algorithm[J].Mountain Research,2026,(1):74-88.[doi:10.16089/j.cnki.1008-2786.000948]
点击复制

基于梯度下降算法优化滑坡运动距离预测模型()

《山地学报》[ISSN:1008-2186/CN:51-1516]

卷:
期数:
2026年第1期
页码:
74-88
栏目:
山地灾害
出版日期:
2026-02-20

文章信息/Info

Title:
Optimization of Landslide Runout Prediction Model Based on Gradient Descent Algorithm
文章编号:
1008-2786-(2026)1-074-15
作者:
周 莹a; 樊晓一a; b*; 夏贵平a; c; 何安江a; 陈家庆a
(西南石油大学 a.土木工程与测绘学院; b.工程安全评估与防护研究院; c.机电工程学院,成都610500)
Author(s):
ZHOU Yinga;  FAN Xiaoyia; b*;  XIA Guipinga; c;  HE Anjianga;  CHEN Jiaqinga
(a. School of Civil Engineering and Geomatic; b. Institute of Engineering Safety Assessment and Protection Research; c. School of Mechatronic Engineering, Southwest Petroleum University, Chengdu 610500, China)
关键词:
坡脚型滑坡;  梯度下降法;  运动距离;  预测模型;  算法优化
Keywords:
slope-toe landslide;  Gradient Descent algorithm;  landslide runout;  prediction model;  algorithm optimization
分类号:
P642; X43
DOI:
10.16089/j.cnki.1008-2786.000948
文献标志码:
A
摘要:
地形差异显著影响滑坡最大运动距离,是滑坡风险评估和灾情预测的关键参数。构建高精度预测模型不仅依赖输入数据的质量、模型泛化的鲁棒性以及迭代过程中参数的无偏估计,还需要有效降低多变量耦合关系所带来的相互干扰。本文提出一种基于梯度下降算法修正的坡脚型滑坡运动距离非线性回归预测模型。该模型有以下特征:(1)该模型通过集成泛化损失和早停机制,优化反向传播框架下的损失梯度计算流程;(2)在数据处理环节,运用孤立森林算法对多源数据整合构建的坡脚型滑坡样本数据库进行异常值检测,以确保数据统计的真实性和准确性,并分别构建基于岩性特征的四组多元非线性回归预测模型;(3)在此基础上,利用梯度下降法集成自适应矩估计(Adaptive Moment Estimation,Adam)优化算法修正模型,提升迭代过程的收敛性,提高计算效率,确定最佳参数。结果表明,采用梯度下降算法对坡脚型滑坡运动距离非线性回归预测模型具有显著优化效果。
Abstract:
Topographic differences significantly affect the maximum runout distance of landslides and are key parameters for landslide risk assessment and geohazard prediction. Constructing a high-precision prediction model depends not only on the quality of input data, the robustness of model generalization, and the unbiased estimation of parameters during iteration, but also requires effectively minimizing the mutual interference caused by multivariate coupling relationships.
In this study, it proposed a nonlinear regression prediction model to estimate the runout distance of slope-toe landslides based on the Gradient Descent algorithm.
(1)The model improved the loss gradient computation process within the backpropagation framework by integrating Generalization Loss and early stopping mechanisms.
(2)In the data processing stage, the Isolation Forest algorithm was employed to detect outliers in a slope-toe landslide database constructed from multi-source data integration, aiming to enhance the authenticity and accuracy of the dataset, then followed by four sets of Multivariate Nonlinear Regression(MNR)models to be constructed separately based on specific geophysical characteristics(lithology).
(3)On this basis, Gradient Descent method was integrated with Adam optimizer and early stopping mechanism, enhancing the convergence of the iterative process, improving computational efficiency, and determining the optimal parameters.
The results verify that incorporating Gradient Descent algorithm significantly improves the predictive performance of the nonlinear regression model by reducing prediction errors and enhancing the stability of landslide runout forecasts.

参考文献/References:

[1] LEGROS F. The mobility of long-runout landslides [J]. Engineering Geology, 2002, 63(3/4): 301-331. DOI: 10.1016/S0013-7952(01)00090-4
[2] 樊晓一, 田述军, 段晓冬, 等. 地形因子对坡脚型地震滑坡运动参数的影响研究[J]. 岩石力学与工程学报, 2014, 33(S2): 4056-4066. [FAN Xiaoyi, TIAN Shujun, DUAN Xiaodong, et al. Study of topography factors influence on motion parameters for seismic slope-toe landslides [J]. Chinese Journal of Rock Mechanics and Engineering, 2014, 33(S2): 4056-4066] DOI: 10.13722/j.cnki.jrme.2014.s2.083
[3] LU Meng, CECCATO F, ZHOU Mingliang, et al. Evaluating the exceedance probability of the runout distance of rainfall-induced landslides using a two-stage FEM-MPM approach [J]. Acta Geotechnica, 2024, 19: 3691-3706. DOI: 10.1007/s11440-023-02160-4
[4] 谢艳芳, 李新坡, 赵曙熙, 等. 基于物质点法的新磨村滑坡动力特性分析[J]. 山地学报, 2018, 36(4): 589-597. [XIE Yanfang, LI Xinpo, ZHAO Shuxi, et al. MPM-based numerical analysis of the kinematic characteristics of Xinmo landslide in Maoxian County, Sichuan, China [J]. Mountain Research, 2018, 36(4): 589-597] DOI: 10.16089/j.cnki.1008-2786.000355
[5] 涂聪, 冯君, 涂正楠, 等. 基于SPH方法的土质滑坡运动距离预测[J]. 自然灾害学报, 2025, 34(2): 89-98. [TU Cong, FENG Jun, TU Zhengnan, et al. Prediction of soil landslide movement distance based on SPH method [J]. Journal of Natural Disasters, 2025, 34(2): 89-98] DOI: 10.13577/j.jnd.2025.0209
[6] MIAO Tiande, LIU Zhongyu, NIU Yonghong, et al. A sliding block model for the runout prediction of high-speed landslides [J]. Candian Geotechnical Journal, 2001, 38(2): 217-226. DOI: 10.1139/cgj-38-2-217
[7] HUNGR O, MCDOUGALL S. Two numerical models for landslide dynamic analysis [J]. Computers and Geosciences, 2009, 35(5): 978-992. DOI: 10.1016/j.cageo.2007.12.003
[8] MA Zhengjing, MEI Gang, PICCIALLI F. Machine learning for landslides prevention: A survey [J]. Neural Computing and Applications, 2021, 33: 10881-10907. DOI: 10.1007/s00521-020-05529-8
[9] 曾渤皓, 王帮圆, 谷丽莹, 等. 基于QR-BP-SVR的坡脚型滑坡运动距离预测[J/OL]. 地质通报, 2026: 1-15 [2025-10-16]. [ZENG Bohao, WANG Bangyuan, GU Liying, et al. Prediction of movement distance for slope-toe landslides based on QR-BP-SVR [J/OL]. Geological Bulletin of China, 2026: 1-15 [2025-10-16] https://link.cnki.net/urlid/11.4648.p.20250929.1054.002. DOI: 10.12097/gbc.2025.04.070
[10] 应欣翰, 吴彩燕, 贾应, 等. 机器学习驱动的滑坡易发性多模型分级优化研究—以四川省理县为例[J]. 现代地质, 2026,40(1): 263-274. [YING Xinhan, WU Caiyan, JIA Ying, et al. Research on multi-model grading optimization of landslide susceptibility driven by machine learning: A case study of Lixian County, Sichuan Province [J]. Geoscience, 2026, 40(1): 263-274] DOI: 10.19657/j.geoscience.1000-8527.2025.069
[11] KRKACˇ M, GAZIBARA S B, ARBANAS , et al. A comparative study of random forests and multiple linear regression in the prediction of landslide velocity [J]. Landslides, 2020, 17: 2515-2531. DOI: 10.1007/s10346-020-01476-6
[12] JU Luyu, XIAO Te, HE Jian, et al. Predicting landslide runout paths using terrain matching-targeted machine learning [J]. Engineering Geology, 2022, 311: 106902. DOI: 10.1016/j.engg eo.2022.106902
[13] SUN Xiaoping, ZENG Peng, LI Tianbin, et al. A Bayesian approach to develop simple run-out distance models: Loess landslides in Heifangtai Terrace, Gansu Province, China [J]. Landslides, 2023, 20: 77-95. DOI: 10.1007/s10346-022-0196 5-w
[14] LIU Zaobao, SHAO Jianfu, XU Weiya, et al. Comparison on landslide nonlinear displacement analysis and prediction with computational intelligence approaches [J]. Landslides, 2014, 11: 889-896. DOI: 10.1007/s10346-013-0443-z
[15] ZHANG Junrong, TANG Huiming, TANNANT D D, et al. A novel model for landslide displacement prediction based on EDR selection and multi-swarm intelligence optimization algorithm [J]. Sensors, 2021,21: 8352. DOI: 10.3390/s21248352
[16] LIMA P, STEGER S, GLADE T, et al. Conventional data-driven landslide susceptibility models may only tell us half of the story: Potential underestimation of landslide impact areas depending on the modeling design [J]. Geomorphology, 2023, 430: 108638. DOI: 10.1016/j.geomorph.2023.108638
[17] ZHANG Huanhuan, DENG Yu, ZHANG Zhen, et al. Power density-dependent friction weakening for long-runout landslides and its revelation of volume and topographical effects [J]. Acta Geotechnica, 2023, 18: 2603-2613. DOI: 10.1007/s11440-022-01644-z
[18] HAO Wenrui. A gradient descent method for solving a system of nonlinear equations [J]. Applied Mathematics Letters, 2021, 112: 106739. DOI: 10.1016/j.aml.2020.106739
[19] KINGMA D P, BA J L. Adam: A method for stochastic optimization [J]. ArXiv Preprint, 2015, arXiv: 1412.6980. DOI: 10.48550/arXiv.1412.6980
[20] 樊晓一, 冷晓玉, 段晓冬. 坡脚型与偏转型地震滑坡运动距离及地形因素作用[J]. 岩土力学, 2015, 36(5): 1380-1388. [FAN Xiaoyi, LENG Xiaoyu, DUAN Xiaodong. Influence of topographical factors on movement distances of toe-type and turning-type landslides triggered by earthquake [J]. Rock and Soil Mechanics, 2015, 36(5): 1380-1388] DOI: 10.16285/j.rsm.2015.05.021
[21] 杨海龙, 裴向军, 樊晓一. 坡脚型滑坡运动特征分析及运动距离预测[J]. 工程地质学报, 2019, 27(6): 1379-1388. [YANG Hailong, PEI Xiangjun, FAN Xiaoyi. Movement characteristics and distance prediction of slope-toe landslides [J]. Journal of Engineering Geology, 2019, 27(6): 1379-1388] DOI: 10.13544/j.cnki.jeg.2018-277
[22] LUO Hongyu, ZHANG Limin, HE Jian, et al. Energy transfer mechanisms in flow-like landslide processes in deep valleys [J]. Engineering Geology, 2022, 308: 106798. DOI: 10.1016/j.enggeo.2022.106798
[23] WANG Jian, HU Xinli, ZHENG Hongchao, et al. Energy transfer mechanisms of mobility alteration in landslide-debris flows controlled by entrainment and runout-path terrain: A case study [J]. Landslides, 2024, 21: 1189-1206. DOI: 10.1007/s10346-024-02221-z
[24] FAN Xiaoyi, TIAN Shujun, ZHANG Youyi. Mass-front velocity of dry granular flows influenced by the angle of the slope to the runout plane and particle size gradation [J]. Journal of Mountain Science, 2016, 13: 234-245. DOI: 10.1007/s11629-014-3396-3
[25] YANG Qingqing, CAI Fei, UGAI K, et al. Some factors affecting mass-front velocity of rapid dry granular flows in a large flume [J]. Engineering Geology, 2011, 122(3/4): 249-260. DOI: 10.1016/j.enggeo.2011.06.006
[26] LIU F T, TING K M, ZHOU Z H. Isolation forest [C]// Rangachar Kasturi. 2008 Eighth IEEE International Conference on Data Mining. Piscataway, NJ: IEEE Computer Society, 2008: 413-422.
[27] 陈晓利, 邓俭良, 冉洪流. 汶川地震滑坡崩塌的空间分布特征[J]. 地震地质, 2011, 33(1): 191-202. [CHEN Xiaoli, DENG Jianliang, RAN Hongliu. Analysis of landslides triggered by Wenchuan earthquake [J]. Seismology and Geology, 2011, 33(1): 191-202] DOI: 10.3969/j.issn.0253-4967.2011.01.018
[28] TAPKIR A. A comprehensive overview of gradient descent and its optimization algorithms [J]. International Advanced Research Journal in Science, Engineering and Technology, 2023,10(11): 37-45. DOI: 10.17148/iarjset.2023.101106
[29] DARE D I, TAOFIKI A A, ONASHOGA A S, et al. An improved gradient descent method for optimization of supervised machine learning problems [J]. International Journal of Computer Applications, 2021, 183(20): 39-45. DOI: 10.5120/ijca2021921564
[30] HECKEL R, YILMAZ F F. Early stopping in deep networks: Double descent and how to eliminate it [J]. ArXiv Preprint, 2020, arXiv: 2007.10099. DOI: 10.48550/arXiv.2007.10099
[31] QIU Haijun, CUI Peng, HU Sheng, et al. Developing empirical relationships to predict loess slide travel distances: A case study on the Loess Plateau in China [J]. Bulletin of Engineering Geology and the Environment, 2018, 77: 1299-1309. DOI: 10.1007/s10064-018-1328-0
[32] GUO Deping, HAMADA M, HE Chuan, et al. An empirical model for landslide travel distance prediction in Wenchuan earthquake area [J]. Landslides, 2014, 11: 281-291. DOI: 10.1007/s10346-013-0444-y

备注/Memo

备注/Memo:
收稿日期(Received date): 2025-10-16; 改回日期(Accepted date): 2026- 02-12
基金项目(Foundation item): 国家自然科学基金(41877524); 四川省自然科学基金面上项目(25NSFSC1319)。[National Natural Science Foundation of China(41877524); General Program of Natural Science Foundation of Sichuan(25NSFSC1319)]
作者简介(Biography): 周莹(2001-),女,四川成都人,硕士研究生,主要研究方向:岩土工程及地质灾害影响和评价。[ZHOU Ying(2001-),female,born in Chengdu, Sichuan Province,M.Sc. candidate, research on geotechnical engineering, evaluation and prediction of geological hazards] E-mail: 1509526916@qq.com
*通讯作者(Corresponding author): 樊晓一(1974-),男,博士,教授,主要研究方向:岩土工程、地质灾害。[FAN Xiaoyi(1974-), male, Ph.D., professor, research on geotechnical engineering and geological disasters] E-mail: fxy@swpu.edu.cn
更新日期/Last Update: 2026-01-30