Research on Inflow Flood Forecasting in Tropical Monsoon Climate Regions Based on LSTM

ZHANG Yonghui; WANG Jian; LIU Taiwen; HUANG Hanliang; WU Chengjun; ZHANG Jingwen; LAI Zhiqiang

doi:10.3969/j.issn.1001-9235.2026.09.007

English Version Journal Login

您当前的位置：

首页 >

文章列表页 >

Research on Inflow Flood Forecasting in Tropical Monsoon Climate Regions Based on LSTM

更新时间：2026-10-10

- Research on Inflow Flood Forecasting in Tropical Monsoon Climate Regions Based on LSTM
- PEARL RIVER Vol. 47, Issue 9, Pages: 66-77(2026)
- 作者机构：
  
  1.华能海南发电股份有限公司东方电厂，海南东方 572600
  2.黄河水利委员会黄河水利科学研究院，河南郑州 450003
  3.中山大学土木工程学院，广东广州 510275
- 作者简介：
- 基金信息：
- DOI：10.3969/j.issn.1001-9235.2026.09.007
  CLC： TV124
- Received：27 November 2025，
  
  Revised：2026-02-07，
  
  Accepted：14 February 2026，
  
  Online First：13 March 2026，
  
  Published：25 September 2026
- 稿件说明：
移动端阅览
张永辉，王健，刘泰文，等.基于LSTM的热带季风气候区入库洪水预报研究［J］.人民珠江，2026，47（9）：66-77.

ZHANG Yonghui,WANG Jian,LIU Taiwen,et al.Research on Inflow Flood Forecasting in Tropical Monsoon Climate Regions Based on LSTM[J].PEARL RIVER,2026,47(09):66-77.
张永辉，王健，刘泰文，等.基于LSTM的热带季风气候区入库洪水预报研究［J］.人民珠江，2026，47（9）：66-77. DOI： 10.3969/j.issn.1001-9235.2026.09.007.

ZHANG Yonghui,WANG Jian,LIU Taiwen,et al.Research on Inflow Flood Forecasting in Tropical Monsoon Climate Regions Based on LSTM[J].PEARL RIVER,2026,47(09):66-77. DOI： 10.3969/j.issn.1001-9235.2026.09.007.

摘要

针对热带季风气候区洪水过程强非线性及传统模型预报精度受限问题，以海南戈枕水库为例，构建基于长短期记忆网络（Long Short-Term Memory，LSTM）的预报模型。研究集成2010—2025年流域内7个单站及面平均降雨、上游出库和区间入库流量数据，经滞后相关分析确定7 h最优窗口。筛选60场洪水按8∶2划分训练与测试集，采用NSE（Nash-Sutcliffe Efficiency）、RMSE（Root Mean Square Error）及合格率评价精度，并引入置换特征重要性解析机制。结果表明，模型测试集平均NSE达0.96；在“摩羯”等典型台风模拟中，NSE超0.96，洪峰及峰现时间误差分别控制在7.5%和±1 h以内。特征分析揭示前期入库与上游出库流量为主导，反映强水文惯性与水力联系；单站降雨差异化贡献验证了模型对产汇流空间异质性的捕捉。该模型有效克服了极端工况下预报不稳定的局限，可支撑精细化防洪调度。

Abstract

Tropical monsoon climate regions present significant challenges for reservoir inflow forecasting due to high-intensity rainfall and rapid runoff generation

which produce strongly nonlinear flood processes that often limit the accuracy of traditional physical models. This study investigates the applicability of a deep learning approach to address these nonlinear forecasting problems

using the Gezhen Reservoir in Hainan Province as the study area. A data-driven forecasting framework based on the Long Short-Term Memory (LSTM) network was developed. A comprehensive 16-year (2010—2025) hourly dataset was compiled

integrating rainfall observations from seven individual stations

basin areal rainfall

upstream outflow from the Daguangba Reservoir

and the interval inflow to the Gezhen Reservoir. Lagged correlation analysis was conducted to quantify the relationship between input variables and target inflow

identifying an optimal time window of 7 hours. A total of 60 historical flood events were selected

with 48 events (80%) used for training and 12 events (20%) for testing. Model performance was evaluated using the Nash-Sutcliffe Efficiency (NSE)

Root Mean Square Error (RMSE)

and a qualification rate based on flood control standards. To interpret the model's predictive mechanism

the Permutation Feature Importance (PFI) method was employed to quantify the contribution of each input factor. Quantitative evaluation demonstrates that the LSTM model achieves high accuracy

with an average NSE of 0.96 on the test set. The model's performance was further validated through simulation of extreme typhoon-induced flood events

exemplified by Typhoon Yagi

where the NSE exceeded 0.96. The relative error of peak discharge was controlled within 7.5%

and the error in peak occurrence time was within ±1 hour. Feature importance analysis identifies antecedent inflow and upstream outflow as the dominant factors

reflecting strong hydrological inertia and hydraulic connectivity. The differentiated contributions of individual rainfall stations further verify the model's capability to capture the spatial heterogeneity of the basin's runoff generation processes. The findings demonstrate that the proposed LSTM model effectively overcomes the instability issues associated with traditional methods under extreme conditions

providing a highly accurate technical approach for reservoir flood forecasting in tropical monsoon regions.

关键词

Keywords

references

黄婧，谭超，陈晓宏，等 . 基于入库洪水识别模型的洪水资源利用潜力评估［J］. 水利学报， 2024 ， 55 （ 11 ）： 1286 - 1297 .

Huang Jing ， Tan Chao ， Chen Xiaohong ， et al . Evaluation of flood resource utilization potential based on identification model for inflow flood ［J］. Journal of Hydraulic Engineering ， 2024 ， 55 （ 11 ）： 1286 - 1297 . （in Chinese）

谢小燕，高熠飞，徐凯莉，等 . 基于气象水文耦合的水库实时洪水预报方法［J］. 中国农村水利水电， 2024 （ 8 ）： 81 - 87，93 .

Xie Xiaoyan ， Gao Yifei ， Xu Kaili ， et al . Real-time reservoir flood forecasting based on meteorological and hydrological coupling ［J］. China Rural Water and Hydropower ， 2024 （ 8 ）： 81 - 87，93 . （in Chinese）

王俊，程海云，郭生练，等 . 智慧流域水文预报技术研究进展与开发前景［J］. 人民长江， 2023 ， 54 （ 8 ）： 1 - 8，59 .

Wang Jun ， Cheng Haiyun ， Guo Shenglian ， et al . Research progress and development prospect of hydrological forecasting technology in smart watersheds ［J］. Yangtze River ， 2023 ， 54 （ 8 ）： 1 - 8，59 .

张梓尚 . 有限训练集下基于物理机制与数据驱动融合模型的径流预报［D］. 咸阳：西北农林科技大学， 2025 .

Zhang Zishang . Runoff prediction based on physical mechanism and data-driven fusion model under limited training set ［D］. Xianyang ： Northwest A&F University ， 2025 . （in Chinese）

Stephanie R C ， Jasmine B D J . Associations between deep learning runoff predictions and hydrogeological conditions in Australia ［J］. Journal of Hydrology ， 2025 ， 651 . DOI： 10.1016/j.jhydrol.2024.132569 http://dx.doi.org/10.1016/j.jhydrol.2024.132569 .

魏兴，陈蒙恩，周育琳，等 . 基于多种混合模型的中长期水文预报研究［J］. 长江科学院院报， 2025 ， 42 （ 10 ）： 24 - 31，53 .

Wei Xing ， Chen Meng'en ， Zhou Yulin ， et al . Mid-long term hydrological forecasting based on multiple hybrid models ［J］. Journal of Yangtze River Scientific Research Institute ， 2025 ， 42 （ 10 ）： 24 - 31，53 . （in Chinese）

陈浩芃，王燕，刘富 . 人工神经网络在径流预测中的应用：以黄河流域兰州水文站为例［J］. 北京印刷学院学报， 2024 ， 32 （ 12 ）： 45 - 54 .

Chen Haopeng ， Wang Yan ， Liu Fu . Application of artificial neural network in runoff prediction： a case study of Lanzhou hydrological station in Yellow River Basin ［J］. Journal of Beijing Institute of Graphic Communication ， 2024 ， 32 （ 12 ）： 45 - 54 . （in Chinese）

鲁志杰，杨丰源，杜昀宸，等 . 基于VMD-LSTM模型的王家坝月径流预测应用研究［J］. 水利信息化， 2025 （ 3 ）： 31 - 34 .

Lu Zhijie ， Yang Fengyuan ， Du Yunchen ， et al . Application research on monthly runoff prediction of Wangjiaba based on VMD-LSTM model ［J］. Water Resources Informatization ， 2025 （ 3 ）： 31 - 34 . （in Chinese）

齐旭，芦钰冰 . 基于LSTM-TCN模型的尚庄炉水库中长期径流预报研究［J］. 水利科学与寒区工程， 2025 ， 8 （ 7 ）： 30 - 34 .

Qi Xu ， Lu Yubing . Research on long-term streamflow prediction in the Shangzhuang Reservoir based on LSTM-TCN model ［J］. Hydro Science and Cold Zone Engineering ， 2025 ， 8 （ 7 ）： 30 - 34 . （in Chinese）

宋海萍，唐一然，党文涛，等 . 改进型LSTM模型在西北干旱区径流模拟中的应用：以祖厉河为例［J］. 人民珠江， 2025 ， 46 （ 8 ）： 50 - 57 .

Song Haiping ， Tang Yiran ， Dang Wentao ， et al . Application of improved LSTM model in runoff simulation in arid region of Northwest China： A Case Study of the Zuli River ［J］. Pearl River ， 2025 ， 46 （ 8 ）： 50 - 57 . （in Chinese）

Reyhaneh H ， Pierre B ， Pierre-andré G ， et al . How can we benefit from regime information to make more effective use of long short-term memory （LSTM） runoff models？［J］. Hydrology Earth System Sciences ， 2022 ， 26 （ 22 ）： 5793 - 5816 .

Hu Feichi ， Yang Qinli ， Yang Junran ， et al . Incorporating multiple grid-based data in CNN-LSTM hybrid model for daily runoff prediction in the source region of the Yellow River Basin ［J］. Journal of Hydrology： Regional Studies ， 2024 ， 51 . DOI： 10.1016/j.ejrh.2023.101652 http://dx.doi.org/10.1016/j.ejrh.2023.101652 .

陶驷骥，李彬权，陈云瑶，等 . 新安江岩溶模型与LSTM模型在洪水模拟中的应用比较［J］. 水电能源科学， 2025 ， 43 （ 10 ）： 6 - 9，55 .

Tao Siji ， Li Binquan ， Chen Yunyao ， et al . Comparison of Karst-Xinanjiang Model and LSTM model in flood simulations ［J］. Water Resources and Power ， 2025 ， 43 （ 10 ）： 6 - 9，55 . （in Chinese）

胡乐怡 . 耦合物理过程模型与机器学习的混合模型在径流预测中的应用研究［D］. 扬州：扬州大学， 2025 . doi: 10.1016/j.advwatres.2025.105177 http://dx.doi.org/10.1016/j.advwatres.2025.105177

Hu Leyi . A novel hybrid framework for combining process-based models with machine learning for streamflow prediction ［D］. Yangzhou ： Yangzhou University ， 2025 . （in Chinese）

Gao Shuai ， Huang Yuefei ， Zhang Shuo ， et al . Short-term runoff prediction with GRU and LSTM networks without requiring time step optimization during sample generation ［J］. Journal of Hydrology ， 2020 ， 589 . DOI： 10.1016/j.jhydrol.2020.125188 http://dx.doi.org/10.1016/j.jhydrol.2020.125188 .

邹艺，朱丽蓉，林尤文，等 . 考虑下垫面影响的海南岛干旱区昌化江流域旱涝时空分布规律［J］. 生态科学， 2024 ， 43 （ 4 ）： 149 - 158 .

Zou Yi ， Zhu Lirong ， Lin Youwen ， et al . Temporal and spatial distribution of drought and flood in Changhua River Basin in arid region of Hainan Island considering the influence of underlying surface ［J］. Ecological Science ， 2024 ， 43 （ 4 ）： 149 - 158 . （in Chinese）

朱丽蓉，刘宇航，叶长青 . 海南岛干旱地区昌化江径流变化及影响因素［J］. 生态科学， 2020 ， 39 （ 1 ）： 183 - 189 .

Zhu Lirong ， Liu Yuhang ， Ye Changqing . Changes in runoff in Changhua River in the arid area of Hainan Island and their influencing factors ［J］. Ecological Science ， 2020 ， 39 （ 1 ）： 183 - 189 . （in Chinese）

张永辉，王健，马俊，等 . 面向多尾水河道的水库发电优化调度研究［J］. 人民珠江， 2025 ， 46 （ 1 ）： 97 - 102 .

Zhang Yonghui ， Wang Jian ， Ma Jun ， et al . Research on optimal power generation scheduling for reservoir in multi-tailed river channels ［J］. Pearl River ， 2025 ， 46 （ 1 ）： 97 - 102 . （in Chinese）

祁永超 . 昌化江调水工程受水区河流水质影响及纳污能力研究［D］. 西安：西安理工大学， 2023 .

Qi Yongchao . Study on the impact of river water quality and pollutant carrying capacity of the Changhua River Water Diversion Project in the water supply area ［D］. Xi'an ： Xi’an University of Technology ， 2023 . （in Chinese）

陈娟，徐琦，曹端祥，等 . 基于多因子多模式集成的中长期径流预测模型［J］. 水科学进展， 2024 ， 35 （ 3 ）： 408 - 419 .

Chen Juan ， Xu Qi ， Cao Duanxiang ， et al . Medium and long-term runoff prediction model based on multi-factor and multi-model integration ［J］. Advances in Water Science ， 2024 ， 35 （ 3 ）： 408 - 419 . （in Chinese）

Views

441

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Research on Drone Swarm Networking Inspection and Intelligent Monitoring for Seawall Engineering

An Intelligent Method for Detecting Leaks in Dams Based on an Improved YOLOv10 Model

Deformation Prediction of Wuxijiang Buttress Dam Based on Whale Optimization Algorithm and Transformer

Intelligent Recognition of Urban Flood Depth Based on Mask RCNN

跨城市内河桥梁工程防洪评价分析

Related Author

Zhang Yonghui

Wang Jian

LIU Taiwen

LIU Gengyuan

SUN Wen

HUANG Xin

YU Zengxin

WANG Liuyi

Related Institution

Huaneng Hainan Power Generation Co., Ltd. Dongfang Power Plant

Guangzhou imapcloud intelligent technology co.,Ltd.

China Water Resources Pearl River Planning Surveying & Designing Co., Ltd.

Technology Consulting (Guangzhou) Co., Ltd., Pearl River Water Resources Commission, Ministry of Water Resources

Pearl River Water Conservancy Comprehensive Technology Center, Pearl River Water Resources Commission, Ministry of Water Resources

AI问答

⁰