GA-BP优化混合模型在强人工干预下复杂流域洪水演进预测中的应用

Application of GA-BP optimized hybrid model in flood routing prediction for complex basins under strong human intervention

摘要

针对强人工干预下流域洪水演进过程非线性耦合强、传统机理模型难以精准刻画的问题，采用一种融合人工干预特征与智能优化的洪水预测方法，以大清河北支北河店断面为对象，选取安格庄出库、北易水及落宝滩流量作为多源输入，构建干预强度比量化指标，并结合遗传算法（genetic algorithm，GA）优化BP神经网络（genetic algorithm-backpropagation，GA-BP），建立t+3 h实时预测模型。结果表明：优化后的GA-BP模型结构为6-5双隐藏层，其预测精度显著优于传统BP模型，测试集均方误差（mean squared error，E_MS）降低40.6%至0.001 1，决定系数R²提升至0.94；敏感性分析证实，水库调度信号（安格庄出库）是预测精度的主导因素；模型在各类工况下均表现稳定，其中在防洪决策的关键泄洪工况下精度最高（R²=0.99）。本研究为强人工干预流域的洪水实时精准预报提供了可解释、高可靠的技术解决方案。

Abstract

The frequency and severity of global flood disasters have dramatically increased due to the combined effects of climate change and human activity, posing serious risks to watershed ecological security and socioeconomic sustainable development. High-precision real-time flood routing prediction is an important component of flood control and disaster mitigation systems. However, traditional hydrological models are unable to meet the urgent demands of real-time forecasting due to complex parameter calibration and low computational efficiency, particularly under rapidly changing rainfall and hydrological conditions. Effectively integrating multi-source heterogeneous inputs and capturing human scheduling signals is a major research challenge for river basins under intense artificial regulation. The goal of this research is to create a hybrid prediction model that overcomes the limitations of traditional data-driven methods, improving flood forecasting accuracy in heavily regulated basins while also providing technical support for flood control operations. This study suggests a hybrid model, known as GA-BP, that combines a genetic algorithm (GA) and a backpropagation neural network (BPNN), with a focus on the heavily intervened Beizhi basin of the Daqing River. To quantitatively characterize the impact of reservoir regulation, a key feature termed the "intervention intensity ratio (η)" was constructed. A composite flood dataset was created by combining data from a rigorously calibrated 1D hydrodynamic model (that simulates design floods at various frequencies) with historical measurements. Model inputs include outflow from Angezhuang Reservoir (human intervention signal), inflow from Beiyishui River (natural runoff), flow at Luobaotan station (upstream signal), and the η feature. The output is the flow at the Beihedian Station. The GA was employed to globally optimize the initial weights and thresholds of the BPNN, overcoming its tendency to converge to local minima. Model performance was evaluated using mean square error, the coefficient of determination ( R ²), and the average relative error of flood peaks. Scenario analyses were conducted under different intervention intensities (strong impoundment, release, and moderate intervention). The results show that the GA-BP model outperforms the conventional BPNN. On the test set, the GA-BP model achieved an E_MS of 0.001 1 and an R ² of 0.94, representing a 40.4% reduction in E_MS and a 2.3% increase in R ² compared to the BPNN (E_MS: 0.002, R ²: 0.92). The average relative error of flood peaks decreased by 26.3%, from 17.58% to 12.96%. Sensitivity analysis revealed the relative importance of input variables: Angezhuang outflow (performance drop of 82.60% upon removal) > Luobaotan flow (61.37% drop) > Beiyishui inflow (44.74% drop), confirming the dominant role of reservoir scheduling. Scenario analysis based on η showed that the model performed robustly across all intervention regimes, achieving notably high precision (R ² = 0.99) during the critical but low-frequency reservoir release scenarios, which are paramount for flood risk management. The GA-BP neural network model with the "intervention intensity ratio" feature designed has proved to be able to overcome the difficulties of flood forecasting in basins with strong artificial regulation. It successfully integrates multi-source signals and captures the complex nonlinear interactions between natural hydrological processes and human scheduling. The model outperforms the standard BPNN in terms of convergence, prediction accuracy, and generalization capability, especially when forecasting high-risk flood release events. This research provides a reliable and efficient technical tool for real-time flood forecasting and refined reservoir operation in heavily regulated basins, contributing to the advancement of smart water management systems.

关键词

强人工干预 / 洪水预测 / 遗传算法 / BP神经网络 / GA-BP混合模型 / 大清河北支流域

Key words

strong artificial intervention / flood prediction / genetic algorithm / BP neural network / GA-BP hybrid model / northern branch of Daqing River basin

参考文献

[1] 刘媛媛, 刘业森, 刘洋, 等. 基于机器学习降雨动态时空特征识别山丘区小流域洪水预报方法研究[J]. 水利学报, 2024, 55(9): 1009-1019. DOI: 10.13243/j.cnki.slxb.20230687.

[2] 李想, 郭丹红, 刘家宏, 等. 京津冀协同发展背景下的县域水资源安全诊断[J]. 水利水电技术（中英文）, 2021, 52(10): 59-71. DOI: 10.13928/j.cnki.wrahe.2021.10.006.

[3] 李国英. 为以中国式现代化全面推进强国建设、民族复兴伟业提供有力的水安全保障: 在2024年全国水利工作会议上的讲话[J]. 水利发展研究, 2024, 24(1): 1-10. DOI: 10.13928/j.cnki.wrdr.2024.03.001.

[4] 翟明慧, 桑国庆, 吴先敏, 等. 基于TCN-Attention-LSTM的大沽河流域场次洪水预测模型[J]. 水电能源科学, 2026, 44(2): 22-27. DOI: 10.20040/j.cnki.1000-7709.2026.20250573.

[5] 甘甜, 郑英, 蒋云钟, 等. 基于DBSCAN-RF洪水分类的洪水预报应用研究[J]. 水利水电技术（中英文）, 2024, 55(3): 77-89. DOI: 10.13928/j.cnki.wrahe.2024.03.007.

[6] Khosravi K, Golkarian A, Tiefenbacher J P. Using optimized deep learning to predict daily streamflow: A comparison to common machine learning algorithms[J]. Water Resources Management, 2022, 36(2): 699-716. DOI: 10.1007/s11269-021-03051-7.

[7] Wang Weilin, Sang Guoqing, Zhao Qiang, et al. Prediction of flash flood peak discharge in hilly areas with ungauged basins based on machine learning[J]. Hydrology Research, 2024, 55(8): 801-814. DOI: 10.2166/nh.2024.004.

[8] 王文川, 杜玉瑾, 和吉, 等. 基于CEEMDAN-VMD-BP模型的月径流量预测研究[J]. 华北水利水电大学学报（自然科学版）, 2023, 44(1): 32-40. DOI: 10.19760/j.ncwu.zk.2023004.

[9] He Sheng, Sang Xuefeng, Yin Junxian, et al. Short-term runoff prediction optimization method based on BGRU-BP and BLSTM-BP neural networks[J]. Water Resources Management, 2023, 37(2): 747-768. DOI: 10.1007/s11269-022-03401-z.

[10] Darabi H, Haghighi A T, Rahmati O, et al. A hybridized model based on neural network and swarm intelligence-grey wolf algorithm for spatial prediction of urban flood-inundation[J]. Journal of Hydrology, 2021, 603: 126854. DOI: 10.1016/j.jhydrol.2021.126854.

[11] 王昶磊, 刘宽, 张扬, 等. 基于SVM的岩体灌浆可灌性分析预测模型[J]. 水利水电技术（中英文）, 2025, 56(增刊1): 372-378. DOI: 10.13928/j.cnki.wrahe.2025.S1.057.

[12] Ibrahim K S M H, Huang Y F, Ahmed A N, et al. A review of the hybrid artificial intelligence and optimization modelling of hydrological streamflow forecasting[J]. Alexandria Engineering Journal, 2022, 61(1): 279-303. DOI: 10.1016/j.aej.2021.04.100.

[13] Xu Wenxin, Chen Jie, Corzo G, et al. Coupling deep learning and physically based hydrological models for monthly streamflow predictions[J]. Water Resources Research, 2024, 60(2): e2023WR035618. DOI: 10.1029/2023WR035618.

[14] Rahimzad M, Moghaddam Nia A, Zolfonoon H, et al. Performance comparison of an LSTM-based deep learning model versus conventional machine learning algorithms for streamflow forecasting[J]. Water Resources Management, 2021, 35(12): 4167-4187. DOI: 10.1007/s11269-021-02937-w.

[15] 雷庆文, 闫磊, 巫晨煜, 等. 基于MK-SVM和时序特征分析的月径流预报模型[J]. 水资源保护, 2024, 40(6): 148-154. DOI: 10.3880/j.issn.1004-6933.2024.06.017.

[16] 张炜, 沈明星, 高成, 等. 集合卡尔曼滤波与随机森林算法在异源遥感降水数据同化融合中的应用[J]. 水电能源科学, 2024, 42(8): 11-16. DOI: 10.20040/j.cnki.1000-7709.2024.20231633.

[17] Vilaseca F, Castro A, Chreties C, et al. Assessing influential rainfall-runoff variables to simulate daily streamflow using random forest[J]. Hydrological Sciences Journal, 2023, 68(12): 1738-1753. DOI: 10.1080/02626667.2023.2232356.

[18] Yao Zhiyuan, Wang Zhaocai, Wang Dangwei, et al. An ensemble CNN-LSTM and GRU adaptive weighting model based improved sparrow search algorithm for predicting runoff using historical meteorological and runoff data as input[J]. Journal of Hydrology, 2023, 625: 129977. DOI: 10.1016/j.jhydrol.2023.129977.

[19] Nearing G, Cohen D, Dube V, et al. Global prediction of extreme floods in ungauged watersheds[J]. Nature, 2024, 627(8004): 559-563. DOI: 10.1038/s41586-024-07145-1.

[20] 申笑萱, 张珂, 张兆安, 等. 水文多模型拆分组配与协同优化研究进展[J]. 水资源保护, 2025, 41(4): 189-196. DOI: 10.3880/j.issn.1004-6933.2025.04.022.

[21] 李法贤, 李航, 吕谋, 等. 基于机理模型与深度学习的小样本洪水预测[J]. 人民黄河, 2026, 48(2): 58-64. DOI: 10.3969/j.issn.1000-1379.2026.02.010.

[22] 刘易文, 李家科, 丁强, 等. 汉江流域安康段降雨径流特征分析及预测[J]. 人民珠江, 2021, 42(6): 59-69. DOI: 10.3969/j.issn.1001-9235.2021.06.010.

[23] 司琪, 金保明, 卢旺铭, 等. 崇阳溪流域PRBP神经网络洪水预报模型研究[J]. 人民珠江, 2025, 46(6): 68-74. DOI: 10.3969/j.issn.1001-9235.2025.06.008.

[24] 袁旭山, 刘京会, 宋珂. 基于BP神经网络的洪涝灾害承灾体脆弱性评估[J]. 人民长江, 2024, 55(2): 26-34. DOI: 10.16232/j.cnki.1001-4179.2024.02.004.

[25] 邹红梅, 朱成涛. 基于LSTM和BP神经网络的水库入库径流中长期预测比较研究[J]. 水文, 2024, 44(4): 27-31. DOI: 10.19797/j.cnki.1000-0852.20230413.

[26] 李东升, 朱奎, 郭艳军, 等. 组合神经网络的城市用水量预测模型研究与应用[J]. 中国水利水电科学研究院学报（中英文）, 2024, 22(6): 579-589. DOI: 10.13244/j.cnki.jiwhr.20240031.

[27] 张忠波, 张双虎, 耿思敏, 等. 大清河流域水工程联合防洪调度研究[J]. 中国水利水电科学研究院学报, 2021, 19(6): 550-556. DOI: 10.13244/j.cnki.jiwhr.20200136.

[28] 马强, 涂泽辉, 李郑淼, 等. 雨带北移影响下大清河北支防洪情势[J]. 南水北调与水利科技（中英文）, 2025, 23(5): 1127-1133. DOI: 10.13476/j.cnki.nsbdqk.2025.0112.

[29] 刘淇, 张小莹, 李琳, 等. 人工神经网络在塔里木河中游流量预测中的应用[J]. 水文, 2025, 45(3): 64-71. DOI: 10.19797/j.cnki.1000-0852.20240138.

[30] Jayaraj T, Raj D M K, Jin D Y, et al. Adaptive momentum-backpropagation algorithm for flood prediction and management in the Internet of Things[J]. Computers, Materials & Continua, 2023, 77(1): 1053-1079. DOI: 10.32604/CMC.2023.038437.

[31] 邱勇, 杨泽文, 周鑫宇, 等. 基于BP神经网络的直角折线堰过流能力预测[J]. 水电能源科学, 2021, 39(3): 74-77. DOI: 10.20040/j.cnki.1000-7709.2021.03.019.

[32] 陈奂良, 李常锁, 高帅, 等. 基于遗传算法优化BP神经网络的岩溶泉水位预测研究[J]. 水资源与水工程学报, 2022, 33(4): 50-57. DOI: 10.11705/j.issn.1672-643X.2022.04.07.

[33] Arante C R H, Sybingco E, Roque A M, et al. Development of a secured IoT-based flood monitoring and forecasting system using genetic-algorithm-based neuro-fuzzy network[J]. Sensors, 2025, 25(13): 3885- 3885. DOI: 10.3390/S25133885.

[34] Afan H A, Wan Mohtar W H M, Aksoy M, et al. A multi-functional genetic algorithm-neural network model for predicting suspended sediment loads[J]. Water Resources Management, 2025, 39(5): 2033-2048. DOI: 10.1007/S11269-024-04054-W.

[35] 李继清, 吴亮, 郑威, 等. 基于遗传算法和Copula函数的流域可供水量计算模型及应用[J]. 中国农村水利水电, 2025(8): 48-54. DOI: 10.12396/znsd.241675.

[36] Samiullah L, Tomoo F. Runoff prediction in small rivers using dynamic parameter fitting with observed rainfall data[J]. Journal of Hydrologic Engineering, 2024, 29(5): 04024023. DOI: 10.1061/JHYEFF.HEENG-6193.

[37] 南宏业, 李翠梅, 王浩, 等. 基于CIWOA-BP和灰色置信区间的银川市需水量预测模型[J]. 人民黄河, 2024, 46(1): 75-78. DOI: 10.3969/j.issn.1000-1379.2024.01.013.

[38] Ren Lyu, Tao Yuan, Liu Jie, et al. Predictions of peak discharge of dam failures based on the combined GA and BP neural networks[J]. Water, 2024, 16(20): 2946. DOI: 10.3390/W16202946.

[39] Shao Yu, Chen Jiarui, Zhang Tuqiao, et al. Advancing rapid urban flood prediction: A spatiotemporal deep learning approach with uneven rainfall and attention mechanism[J]. Journal of Hydroinformatics, 2024, 26(6): 1409-1424. DOI: 10.2166/hydro.2024.024.

[40] Hu Yaosheng, Tang Ming, Ma Shuaitao, et al. A rapid prediction method for key information of the urban flood control engineering system based on machine learning: An empirical study of the Wusha River basin[J]. Water, 2025, 17(6): 784- 784. DOI: 10.3390/W17060784.

基金资助

国家重点研发计划项目(2023YFC3006503)

河北省自然科学基金项目(E2024402142)

河北省省级水利科技计划项目(HBSL2025-04)