基于级联物理信息神经网络的库岸边坡瞬态渗流模拟

doi:10.3799/dqkx.2026.170

基于级联物理信息神经网络的库岸边坡瞬态渗流模拟

doi: 10.3799/dqkx.2026.170

甘龙厚^1,2,
彭铭^1,2, ,

1. 同济大学土木工程防灾减灾全国重点实验室, 上海 200092;
2. 同济大学土木工程学院, 上海 200092

基金项目:

国家自然科学基金-联合基金重点项目(U23A2044)。

详细信息

作者简介:
甘龙厚(2001-),男,博士研究生,从事库岸滑坡灾害机理与风险评估研究。E-mail:2310487@tongji.edu.cn,ORCID:0009-0005-9269-0997

通讯作者:
彭铭(1981-),男,教授,从事地质灾害机理与防治研究。E-mail:pengming@tongji.edu.cn

中图分类号: P642
计量
- 文章访问数: 71
- HTML全文浏览量: 0
- PDF下载量: 13
- 被引次数: 0
出版历程
- 收稿日期: 2026-03-02
- 网络出版日期: 2026-06-29

Unsaturated transient seepage simulation of reservoir bank slope using a cascading physics-informed neural network(CPINN)

Longhou Gan^1,2,
Ming Peng^{1,2
, ,}

1. State Key Laboratory of Disaster Reduction in Civil Engineering, Tongji University, Shanghai 200092, China;
2. College of Civil Engineering, Tongji University, Shanghai 200092, China

摘要

摘要: 降雨与库水位作为库岸滑坡的主要诱因，通过改变边坡内渗流行为，进而影响边坡稳定性。因此，准确表征库岸边坡内渗流场的时空演化特征是开展稳定性分析的关键。然而，在降雨与库水位的联合作用下，边坡渗流过程受动态渗透边界控制，呈现显著的非平稳特征（如边界条件突变、水头响应滞后等），给传统数值方法的稳定求解带来了挑战。针对上述问题，本文提出一种级联物理信息神经网络模型（Cascading physics-informed neural network，CPINN）。该方法采用“分段建模-顺序训练-状态传递”的级联策略，分阶段学习由动态边界引起的非平稳物理过程。将全时间域按照渗透边界或入渗方式在时间上的阶段性变化划分为多个子时段，并为每个时段分别构建子 PINN 模型；训练时引入时间顺序推进机制，在保证物理约束一致性的前提下，实现各子模型之间的连续衔接与信息传递；预测时根据目标区段调用对应子模型，使框架既能捕捉水头随时间变化的全局趋势，又能反映不同渗透边界下的局部瞬态响应。结果表明，级联PINN有效解决了动态边界问题、显著降低了求解难度，并实现降雨与水位联合作用下库岸边坡瞬态渗流的准确模拟，为此类具有动态边界变化的复杂渗流问题提供了一种高效稳定的建模策略。
- 库岸边坡 /
- 非饱和瞬态渗流 /
- 数值模拟 /
- 深度学习 /
- 物理信息神经网络
Abstract: Rainfall and reservoir water-level fluctuations influence reservoir bank slope stability through their alteration of seepage behavior within the slope. Therefore, accurately characterizing the spatiotemporal evolution of the seepage field is essential for slope stability analysis. However, under the combined effects of rainfall infiltration and reservoir water-level variations, the seepage process is governed by dynamic hydraulic boundary conditions and exhibits pronounced non-stationary characteristics (e.g., abrupt changes in boundary conditions and lag in hydraulic head response), which pose significant challenges to the stable solution of conventional numerical simulation methods. To address this issue, this study proposes a cascading physics-informed neural network (CPINN) framework. The proposed method adopts a cascading strategy of “segmented modeling-sequential training-state transfer” to learn non-stationary physical processes induced by dynamically varying boundary conditions in a stage-wise manner. Specifically, the entire time domain is divided into multiple sub-intervals according to the temporal evolution of seepage boundary conditions or infiltration patterns, and a dedicated sub-PINN model is constructed for each interval. During training, a sequential time-marching mechanism is introduced to ensure continuous information transfer and consistent physical constraints among successive sub-models. During prediction, the corresponding sub-model is activated based on the target time interval, enabling the framework to capture both the global temporal evolution of hydraulic head and the local transient responses under different seepage boundary conditions. The results demonstrate that the proposed CPINN effectively addresses dynamic boundary conditions, significantly reduces the difficulty of solving, and achieves high-accuracy simulation of transient seepage in reservoir bank slopes under the combined effects of rainfall and reservoir water-level fluctuations. This approach provides an efficient and robust modeling strategy for complex seepage problems involving dynamically evolving boundary conditions.
- Reservoir bank slope /
- Unsaturated transient seepage /
- Numerical simulation /
- Deep learning /
- Physics-informed neural network(PINN)

HTML全文

参考文献(34)

[1]	Chen, Y., Xu, Y., Wang, L. and Li, T., 2023. Modeling Water Flow in Unsaturated Soils through Physics-Informed Neural Network with Principled Loss Function. Computers and Geotechnics, 161: 105546. https://doi.org/10.1016/j.compgeo.2023.105546
[2]	Deng Zhiping, Wan Minhao, Pan Min, et al. Slope Reliability Analysis Considering Degradation and Spatial Variability of Subsidence Zone Parameters. Chinese Journal of Earth Science, 2026, 51(2): 547-559 (in Chinese with English abstract).
[3]	Gao, Y., Qian, L., Yao, T., Mo, Z., Zhang, J., Zhang, R., Liu, E., Li, Y. and Yaseen, Z.M., 2023. An Improved Physics-Informed Neural Network Algorithm for Predicting the Phreatic Line of Seepage. Advances in Civil Engineering, 2023: 1-11. https://doi.org/10.1155/2023/5499645
[4]	Han Jia-ming, Dong Zhao, Su San-qing, et al, 2023. Analytical Solution of Rainfall Infiltration in Homogeneous Unsaturated Slope and Its Application in Loess Slope. Chinese Journal of Rock and Soil Mechanics, 44(01): 241-250 (in Chinese with English abstract).
[5]	Jiang Wen-hao, Wang Hao, Liao Guang-zhi, Chen Bin-hua, 2025. Analytical Solutions for One-Dimensional Transient Seepage of Water in the Two-Layered Unsaturated Soils under Time-Varying Rainfall Conditions. Chinese Journal of Rock and Soil Mechanics, 46(09): 1-17 (in Chinese with English abstract).
[6]	Lan, P., Su, J. and Zhang, S., 2024. Surrogate Modeling for Unsaturated Infiltration via the Physics and Equality-Constrained Artificial Neural Networks. Journal of Rock Mechanics and Geotechnical Engineering, 16(6): 2282-2295. https://doi.org/10.1016/j.jrmge.2023.09.014
[7]	Li Changdong, Tan Qinwen, 2022. Can the Physical Sliding of Hydrodynamic-Driven Landslide Be Predicted? Chinese Journal of Earth Science, 47(10): 3908-3910 (in Chinese with English abstract).
[8]	Li, X., Li, X., Wu, Y., Wu, L. and Yue, Z., 2022. Selection Criteria of Mesh Size and Time Step in FEM Analysis of Highly Nonlinear Unsaturated Seepage Process. Computers and Geotechnics, 146: 104712. https://doi.org/10.1016/j.compgeo.2022.104712
[9]	Lv, S., Li, D., Zha, W., Shen, L. and Xing, Y., 2024. Solving Seepage Equation Using Physics-Informed Residual Network without Labeled Data. Computer Methods in Applied Mechanics and Engineering, 418: 116563. https://doi.org/10.1016/j.cma.2023.116563
[10]	Ng, C.W.W., Usman, M. and Guo, H., 2023. Spatiotemporal Pore-Water Pressure Prediction Using Multi-Input Long Short-Term Memory. Engineering Geology, 322: 107194. https://doi.org/10.1016/j.enggeo.2023.107194
[11]	Orland, E., Roering, J.J., Thomas, M.A. and Mirus, B.B., 2020. Deep Learning as a Tool to Forecast Hydrologic Response for Landslide-Prone Hillslopes. Geophysical Research Letters, 47(16). https://doi.org/10.1029/2020gl088731
[12]	Peng Ming, Wang Yue, Ma Chenyi, Shi Zhenming, Zhu Yan, 2025. Review of Risk Assessment and Prevention for Valley Landslide Disaster Chains. Chinese journal of Earth Science, 50(10): 3723-3760 (in Chinese with English abstract).
[13]	Rong Guan, Zhang Wei, Zhou Chuang-bing, 2005. Numerical Analysis of Saturated-Unsaturated Seepage Problem of Rock Slope under Rainfall Infiltration. Chinese Journal of Rock and Soil Mechanics, 26(10): 24-29 (in Chinese with English abstract).
[14]	Shen Yanshen, Zeng Xin, Chen Feifei, et al, 2026. Experimental Study on the Effects of Different Rainfall Patterns on Deformation and Failure of Bedding Rock Slopes with Weak Interlayers. Chinese Journal of Bulletin of Geological Science and Technology, 45(2): 180-195 (in Chinese with English abstract).
[15]	Sun, Y., Li, D., Jiao, L., Liu, Z., Yang, Y., Lyu, J. and Lin, S., 2024. Simulation of Saturated-Unsaturated Seepage Problems via the Virtual Element Method. Computers and Geotechnics, 171: 106326. https://doi.org/10.1016/j.compgeo.2024.106326
[16]	Tang Huiming, Li Changdong, Gong Wenping, et al, 2022. Fundamental Attribute and Research Approach of Landslide Evolution. Chinese Journal of Earth Science, 47(12): 4596-4608 (in Chinese with English abstract).
[17]	Tang, H., Wasowski, J. and Juang, C.H., 2019. Geohazards in the Three Gorges Reservoir Area, China - Lessons Learned from Decades of Research. Engineering Geology, 261: 105267. https://doi.org/ 10.1016/j.enggeo.2019.105267
[18]	Wei, X., Zhang, L., Yang, H.-Q., Zhang, L. and Yao, Y.-P., 2021. Machine Learning for Pore-Water Pressure Time-Series Prediction: Application of Recurrent Neural Networks. Geoscience Frontiers, 12(1): 453-467. https://doi.org/10.1016/j.gsf.2020.04.011
[19]	Wu Yiping, Yi Yucong, Li Shuo, et al, 2026. Risk Assessment of Rangduchang North Landslide in Wanzhou District, Three Gorges Reservoir area. Chinese Journal of Bulletin of Geological Science and Technology, 45(3): 1-10 (in Chinese with English abstract).
[20]	Xiong, X., Shi, Z., Xiong, Y., Peng, M., Ma, X. and Zhang, F., 2019. Unsaturated Slope Stability around the Three Gorges Reservoir under Various Combinations of Rainfall and Water Level Fluctuation. Engineering Geology, 261: 105231. https://doi.org/10.1016/j.enggeo.2019.105231
[21]	Yang, H.-Q., Shi, C. and Zhang, L., 2025. Ensemble Learning of Soil-Water Characteristic Curve for Unsaturated Seepage Using Physics-Informed Neural Networks. Soils and Foundations, 65(1): 101556. https://doi.org/10.1016/j.sandf.2024.101556
[22]	Zhan Liang-tong, Jia Guan-wei, Chen Yun-min, FREDLUND D G, 2010. Analytical Solution for Rainfall Infiltration into Infinite Long Slopes Considering Properties of Unsaturated Soil. Chinese Journal of Geotechnical Engineering, 32(08): 1214-1220 (in Chinese with English abstract).
[23]	Zhu, S.R., Wu, L.Z., Shen, Z.H. and Huang, R.Q., 2019. An Improved Iteration Method for the Numerical Solution of Groundwater Flow in Unsaturated Soils. Computers and Geotechnics, 114: 103113. https://doi.org/10.1016/j.compgeo.2019.103113
[24]	Zhu, Y., Qiu, H., Liu, Z., Ye, B., Tang, B., Li, Y. and Kamp, U., 2024. Rainfall and Water Level Fluctuations Dominated the Landslide Deformation at Baihetan Reservoir, China. Journal of Hydrology, 642: 131871. https://doi.org/10.1016/j.jhydrol.2024.131871
[25]	邓志平, 万旻昊, 潘敏, 等, 2026. 考虑消落带参数退化与空间变异的边坡可靠度分析. 地球科学, 51(2): 547-559.
[26]	江文豪, 王浩, 廖光志, 等, 2025. 时变降雨场景下双层非饱和土中水分一维瞬态渗流解析解. 岩土力学, 46(09): 1-17.
[27]	韩佳明, 董照, 苏三庆, 等, 2023. 均质非饱和边坡降雨入渗解析解及在黄土边坡的应用. 岩土力学, 44(01): 241-250.
[28]	李长冬, 谭钦文, 2022. 动水驱动型滑坡物理启滑能够预测吗? 地球科学, 47(10): 3908-3910.
[29]	彭铭, 王悦, 马晨议, 石振明, 朱艳, 2025. 河谷滑坡灾害链风险评估及防控研究进展. 地球科学, 50(10): 3723-3760.
[30]	荣冠, 张伟, 周创兵, 2005. 降雨入渗条件下边坡岩体饱和非饱和渗流计算.岩土力学, 26(10): 24-29.
[31]	沈颜参, 曾鑫, 陈菲菲, 等, 2026. 不同降雨雨型对含软弱夹层顺层岩质边坡变形破坏影响的试验. 地质科技通报, 45(2): 180-195.
[32]	唐辉明, 李长冬, 龚文平, 等, 2022.滑坡演化的基本属性与研究途径. 地球科学, 47(12): 4596-4608.
[33]	吴益平, 易羽聪, 李硕, 等, 2026. 三峡库区万州区瀼渡场北滑坡灾害风险评价. 地质科技通报, 45(3): 1-10.
[34]	詹良通, 贾官伟, 陈云敏, FREDLUND D G, 2010. 考虑土体非饱和特性的无限长斜坡降雨入渗解析解. 岩土工程学报, 32(08): 1214-1220.

施引文献

资源附件(0)

访问统计

点击查看大图

计量

文章访问数: 71
HTML全文浏览量: 0
PDF下载量: 13
被引次数: 0

姓名
邮箱
手机号码
标题
留言内容
验证码

留言板

基于级联物理信息神经网络的库岸边坡瞬态渗流模拟

doi: 10.3799/dqkx.2026.170

作者简介:
甘龙厚(2001-),男,博士研究生,从事库岸滑坡灾害机理与风险评估研究。E-mail:2310487@tongji.edu.cn,ORCID:0009-0005-9269-0997

通讯作者:
彭铭(1981-),男,教授,从事地质灾害机理与防治研究。E-mail:pengming@tongji.edu.cn

计量

Unsaturated transient seepage simulation of reservoir bank slope using a cascading physics-informed neural network(CPINN)

计量

目录

留言板

基于级联物理信息神经网络的库岸边坡瞬态渗流模拟

doi: 10.3799/dqkx.2026.170

作者简介: 甘龙厚(2001-),男,博士研究生,从事库岸滑坡灾害机理与风险评估研究。E-mail:2310487@tongji.edu.cn,ORCID:0009-0005-9269-0997

通讯作者: 彭铭(1981-),男,教授,从事地质灾害机理与防治研究。E-mail:pengming@tongji.edu.cn

计量

出版历程

Unsaturated transient seepage simulation of reservoir bank slope using a cascading physics-informed neural network(CPINN)

计量

出版历程

目录

作者简介:
甘龙厚(2001-),男,博士研究生,从事库岸滑坡灾害机理与风险评估研究。E-mail:2310487@tongji.edu.cn,ORCID:0009-0005-9269-0997

通讯作者:
彭铭(1981-),男,教授,从事地质灾害机理与防治研究。E-mail:pengming@tongji.edu.cn