[1]万俞,别如雪.基于VMD-BiLSTM-AM的水库滑坡变形预测研究[J].江苏水利,2026,(08):34-39.
WAN Yu,BIE Ruxue.Research on reservoir landslide deformation prediction based on VMD-BiLSTM-AM[J].JIANGSU WATER RESOURCES,2026,(08):34-39.
点击复制
基于VMD-BiLSTM-AM的水库滑坡变形预测研究(
)
《江苏水利》[ISSN:1006-6977/CN:61-1281/TN]
- 卷:
-
- 期数:
-
2026年08期
- 页码:
-
34-39
- 栏目:
-
水利信息化
- 出版日期:
-
2026-08-01
文章信息/Info
- Title:
-
Research on reservoir landslide deformation prediction based on VMD-BiLSTM-AM
- 文章编号:
-
1007-7839(2026)08-0038-0006
- 作者:
-
万俞; 别如雪
-
(江苏省工程勘测研究院有限责任公司,江苏 扬州 225000)
- Author(s):
-
WAN Yu; BIE Ruxue
-
(Jiangsu Province Engineering Investigation and Research Institute Co., Ltd., Yangzhou 225000, China)
-
- 关键词:
-
VMD-BiLSTM-AM; 水库滑坡; 变形预测; GNSS
- Keywords:
-
VMD-BiLSTM-AM; reservoir landslide; deformation prediction; GNSS
- 分类号:
-
TV697.23
- 文献标志码:
-
B
- 摘要:
-
为解决水库滑坡变形预测中监测信号非线性所导致的预测精度低的问题,以仑山水库库区滑坡体为研究对象,提出了一种基于变分模态分解(VMD)、双向长短时记忆网络(BiLSTM)和注意力机制(AM)的组合变形预测模型。利用VMD将GNSS系统获得的原始位移序列分解为趋势、周期以及噪声成分,有效去除噪声后生成重构序列;利用BiLSTM从处理后的重组序列中双向提取时变特征,引入注意力机制对关键时间步的信息进行自适应加权,最后完成未来不同时间的变形预测。结果表明,VMD-BiLSTM-AM模型预测准确率远高于LSTM-AM、BiLSTM模型,未来3 d的预测均方根误差(RMSE)为0.82 mm,平均绝对误差(MAE)为0.64 mm,决定系数(R2)为0.973,其预测精度与残差集中度均优于传统的BiLSTM及LSTM-AM模型。
- Abstract:
-
To solve the problem of low prediction accuracy caused by non-linear monitoring signals in reservoir landslide deformation prediction, taking the landslide bodies in the Lunshan Reservoir area as the research object, a combined deformation prediction model based on Variational Mode Decomposition (VMD), Bidirectional Long Short-Term Memory Network (BiLSTM), and Attention Mechanism (AM) is proposed. The original displacement series obtained from the GNSS system is decomposed into trend, periodic, and noise components using VMD, effectively removing noise to generate a reconstructed sequence. BiLSTM is then employed to bidirectionally extract time-varying features from the processed reconstructed sequence, while the attention mechanism is introduced to adaptively weight key time steps. Finally, deformation predictions for different future time periods are completed. The results show that the VMD-BiLSTM-AM model achieves significantly higher prediction accuracy than the LSTM-AM and BiLSTM models, with a root mean square error (RMSE) of 0.82 mm and a mean absolute error (MAE) of 0.64 mm for the 3-day forecast, along with a coefficient of determination (R2) of 0.973. Its prediction accuracy and residual concentration outperform traditional BiLSTM and LSTM-AM models.
参考文献/References:
[1]唐勇. GNSS在水库滑坡体变形监测中的应用[J]. 云南水力发电,2025,41(7):70-72.
[2]张乔峰,汤明高,周剑,等. 白鹤滩水库滑坡变形及对库水变化的敏感性分析[J]. 水文地质工程地质,2026,53(2):213-222.
[3]王冬,闫小龙,申永斌. 基于数值模拟的水库滑坡变形特征预报研究[J]. 水科学与工程技术,2023(3):65-68.
[4]郭子琦,张秋昭,安亚杰,等. 时序InSAR技术水库滑坡隐患区识别的应用研究[J]. 地理空间信息,2025,23(8):94-97.
[5]王石,胥锡茂,黄杰,等. 基于SSA-CNN-BiLSTM-AM和KDE的滑坡位移混合点-区间预测方法[J]. 工程地质学报,2025,33(6):2187-2198.
[6]何清,李丽琳,林子安. 基于FEEMD-GRU-FC模型的滑坡位移预测[J]. 人民长江,2024,55(7):108-114.
备注/Memo
- 备注/Memo:
-
收稿日期:2026-03-26
作者简介:万俞(1994—),男,工程师,本科,主要从事水利水电工程工作。E-mail: quanmai2985@163.com
更新日期/Last Update:
2026-08-01