Page 47 - 2025年第56卷第7期

［ 21］ 王晓玲，李克，张宗亮，等 . 耦合 ALO-LSTM 和特征注意力机制的土石坝渗压预测模型［J］ 水利学报 ，
.
2022，53（4）：403-412.
［ 22］ 郑书闽，颜建国，郭鹏程，等 . 基于 VMD 及深度学习的供水管道小尺度泄漏检测研究［J］ 水利学报，2024，
.
55（8）：999-1008.
［ 23］ 陈鹏，吴一凡，蔡爽，等 . 基于自编码压缩与多尺度特征提取的抽水蓄能机组劣化趋势评估与预测［J］ 水利
.
学报，2022，53（6）：747-756.
［ 24］ NIU Z Y，ZHONG G Q，YU H. A review on the attention mechanism of deep learning［J］ Neurocomputing，2021，
.
452（10）：48-62.
［ 25］ 崔震，郭生练，王俊，等 . 基于混合深度学习模型的洪水过程概率预报研究［J］ 水利学报，2023，54（8）：
.
889-897，909.
［ 26］ 吕菲，钟登华，余佳，等 . 迁移学习框架下高心墙堆石坝施工仿真参数 IGOA-MLP 动态预测模型［J］ 水利
.
学报，2023，54（10）：1151-1162.
［ 27］ XUE J K，SHEN B. A novel swarm intelligence optimization approach：Sparrow search algorithm［J］ Systems Sci⁃
.
ence & Control Engineering，2020，8（1）：22-34.
.
［ 28］ KUSCHNIG N， VASHOLD L. BVAR： Bayesian vector autoregressions with hierarchical prior selection in R［J］
Journal of Statistical Software，2019，100（14）：i14.
［ 29］ HAMILTON J D. Time Series Analysis［M］ Princeton：Princeton University Press，2020.
.
［ 30］ PAPARODITIS E， POLITIS D N. The asymptotic size and power of the augmented dickey-fuller test for a unit root
［J］ Econometric Reviews，2016，37（9）：955-973.
.
［ 31］ BADSHAH W，BULUT M. Model selection procedures in bounds test of cointegration：Theoretical comparison and
empirical evidence［J］ Economies，2020，8（2）：49.
.

Research on the predictive model for seepage in concrete dams considering the hysteresis
effects of reservoir water level and temperature variations

1，2 1 1 1 1 3
CHEN Xudong ，LAN Tingting ，HU Shaowei ，XU Ying ，GUO Jinjun ，GU Chongshi

（1. Zhengzhou University，School of Water Conservancy and Transportation，Zhengzhou 450001，China；

2. Research Center on National Dam Safety Engineering Technology，Wuhan 430010，China；

3. The National Key Laboratory of Water Disaster Prevention，Nanjing 210098，China）

Abstract：Seepage behavior is a comprehensive reflection of the interaction between external environmental loads，

such as reservoir water level and temperature，and the internal anti-seepage and drainage structures. However，there

is no effective qualification method at present for the hysteresis effect of reservoir water level and temperature change

on seepage. This study aims to explore the law of hysteresis effect，develop a quantitative expression of the hysteresis
effect，and establish a seepage prediction model accordingly. The Bayesian Vector Autoregression （BVAR）model

was firstly used to analyze the hysteresis process of reservoir water level and temperature on seepage flow，quantita⁃
tively representing the components of reservoir water level and temperature influence. Secondly，to effectively charac⁃

terize the non-linear mapping relationship between seepage and influencing factors，the Attention Mechanism （AM）

was used to dynamically adjust influence weights of seepage input factors，and the Bidirectional Gated Recurrent

Unit （BiGRU）was strengthened to screen key information. The Sparrow Search Algorithm （SSA）was introduced to

improve global search and adaptive performance，establishing AM-SSA-BiGRU model for seepage prediction of con⁃
crete dams. The case study demonstrates that the BVAR method can reflect the hysteresis process of reservoir water
level and temperature effects on seepage. The AM-SSA-BiGRU prediction model effectively captures the seepage
trend with high accuracy and robustness，which provides a novel approach for a deeper understanding of the evolution

of seepage patterns and performance prediction of concrete dams.

Keywords：concrete dams；seepage behavior；hysteresis effect；AM-SSA-BiGRU prediction model；Bayesian

vector autoregression
（责任编辑：韩 昆）
— 873 —