基于多种机器学习模型的大汶河流域径流模拟与对比分析

Runoff Simulation and Comparative Analysis of the Dawen River Basin Based on Multiple Machine Learning Models

摘要

为探究不同机器学习模型在月径流模拟中的适用性与模拟效果，本文以大汶河流域为研究区域，通过Mann-Kendall检验和Sen’s slope方法分析了1961-2019年水文气象要素的变化趋势、突变特征和相关关系，构建了基于卷积神经网络（CNN）、长短期记忆网络（LSTM）和Transformer的月尺度径流模拟模型，并系统对比了三种模型的模拟效果。结果表明：研究时段内流域降水和径流均呈现出不显著下降趋势，气温显著上升，蒸散发、气压和日照时长显著下降；其中，降水发生多次突变，年际变化较大；径流在1966年和2013年发生两次突变，受人类活动影响较大；在月径流的模拟中Transformer表现最佳，测试期NSE和R分别达到0.74和0.87；此外，Transformer具备较强的稳定性与泛化能力，能够在人类活动干扰背景下有效提取不同时期气象要素与径流之间的内在一致性关系，从而实现径流过程的精确模拟。

Abstract

To explore the applicability and simulation performance of different machine learning models in monthly runoff simulation, this paper takes the Dawen River Basin as the study area. Using the Mann-Kendall test and Sen’s slope method, it analyzes the trends, abrupt changes, and correlations of hydrometeorological variables from 1961 to 2019. It constructs monthly streamflow simulation models based on the Convolutional Neural Network (CNN), Long Short-Term Memory network (LSTM), and Transformer, and systematically compares the simulation results of the three models. The results indicate that during the study period, both precipitation and streamflow exhibited no significant decline, while temperature increased significantly, and evapotranspiration, air pressure, and sunshine duration decreased markedly. Precipitation experienced multiple abrupt changes with strong interannual variability; streamflow exhibited two abrupt shifts in 1966 and 2013, showing a strong influence of human activities. In monthly streamflow simulation, the Transformer performed best, achieving NSE and R values of 0.74 and 0.87, respectively, during the testing period. Furthermore, the Transformer demonstrated strong stability and generalization capability, effectively capturing the intrinsic consistency between meteorological variables and streamflow under the influence of human activities, thereby enabling accurate simulation of the streamflow process.

关键词

径流模拟 / 机器学习 / 大汶河流域 / 气候变化

Key words

Runoff simulation / machine learning / Dawen River Basin / climate change

参考文献

[1] 杨大文, 徐宗学, 李 哲, 等 . 水文学研究进展与展望[J]. 地理科学进展, 2018, 37(01): 36-45.

[2] 宋晓猛, 张建云, 占车生, 等. 气候变化和人类活动对水文循环影响研究进展[J]. 水利学报, 2013, 44(07): 779-790.

[3] 徐宗学, 姜 瑶. 变化环境下的径流演变与影响研究:回顾与展望[J]. 水利水运工程学报, 2022(01): 9-18.

[4] Piao S, Ciais P, Huang Y, et al. The impacts of climate change on water resources and agriculture in China[J]. Nature, 2010, 467(7311): 43-51.

[5] Xu Y, Zhang X, Wang X, et al. Propagation from meteorological drought to hydrological drought under the impact of human activities:A case study in northern China[J]. Journal of Hydrology, 2019, 579: 124147.

[6] 周国逸,李 琳, 方雪纯, 等. 近 60 a 来全球河川径流量变化趋势及影响因素分析[J]. 大气科学学报, 2021, 43(06): 1076-1085.

[7] 牛凯杰,吴 为, 刘庚, 等. 永定河流域水文响应变化归因与驱动因素研究[J]. 水资源保护, 2026, 42(02): 140-150.

[8] 张 力, 王红瑞, 郭琲楠, 等. 基于时序分解与机器学习的非平稳径流序列集成模型与应用[J]. 水科学进展, 2023, 34(01): 42-52.

[9] Kratzert F, Klotz D, Brenner C, et al. Rainfall-runoff modelling using long short-term memory (LSTM) networks[J]. Hydrology and Earth System Sciences, 2018, 22(11): 6005-6022.

[10] Xu W, Chen J, Zhang X J, et al. A framework of integrating heterogeneous data sources for monthly streamflow prediction using a state-of-the-art deep learning model[J]. Journal of Hydrology, 2022, 614: 128599.

[11] Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need[J]. Advances in Neural Information Processing Systems, 2017, 30.

[12] Yin H, Guo Z, Zhang X, et al. RR-Former: Rainfall-runoff modeling based on Transformer[J]. Journal of Hydrology, 2022, 609: 127781.

[13] 夏 军, 占车生, 曾思栋, 等. 长江模拟器的理论方法与实践探索[J]. 水利学报, 2022, 53(05): 505-514.

[14] 何振芳, 牟婷婷, 郭庆春, 等. 1979-2019年大汶河流域湿地时空演变与分异研究[J]. 水资源保护, 2024, 40(02): 134-140.

[15] 雒艳玲, 乔新贺, 徐植信, 等. 大汶河流域水文要素时空分布特征研究[J]. 山东农业大学学报(自然科学版), 2024, 55(04): 613-626.

[16] 徐 珊,李 环,陈 新, 等. 基于遥感数据的大汶河流域济南段水体面积变化分析及生态流量评价[J]. 中国农村水利水电, 2024(12): 52-59+ 73.

[17] 廖安栋. 大汶河流域洪水预报及参数不确定性分析[D]. 北京: 华北水利水电大学, 2024.

[18] Sen P K. Estimates of the regression coefficient based on Kendall's tau[J]. Journal of the American Statistical Association, 1968, 63(324): 1379-1389.

[19] Lecun Y, Boser B, Denker J S, et al. Backpropagation applied to handwritten zip code recognition[J]. Neural Computation, 1989, 1(4): 541-551.

[20] 刘建伟, 宋志妍. 循环神经网络研究综述[J]. 控制与决策, 2022, 37(11): 2753-2768.

[21] Yin H,Zhu W,Zhang X, et al. Runoff predictions in new-gauged basins using two transformer-based models[J]. Journal of Hydrology, 2023, 622: 129684.

[22] 杨旭洋. 变化环境下流域径流演变特征及归因分析研究[D]. 西安: 西安理工大学, 2023.

[23] 杨彦武, 于 强,王 靖. 近40年华北及华东局部主要气候资源要素的时空变异性[J]. 资源科学, 2004(04): 45-50.

[24] 任国玉, 战云健, 任玉玉, 等. 中国大陆降水时空变异规律——I. 气候学特征[J]. 水科学进展, 2015, 26(03): 299-310.

[25] 张建云, 张成凤, 鲍振鑫, 等. 黄淮海流域植被覆盖变化对径流的影响[J]. 水科学进展, 2021, 32(06): 813-823.

[26] 刘昌明, 王恺文,王 冠, 等. 1956-2016 年黄河流域径流及其影响因素的变化分析[J]. 人民黄河, 2022, 44(09): 1-5+16.

[27] 刘 园,王 颖, 杨晓光. 华北平原参考作物蒸散量变化特征及气候影响因素[J]. 生态学报, 2010(04): 923-932.

[28] 李勇刚,赵 龙, 李建新, 等. 黄河下游大汶河流域水文要素演变特征研究[J]. 山东农业大学学报(自然科学版), 2023, 54(01): 104-111.

[29] 郭坤荣. 大汶河生态健康评价研究[D]. 济南: 山东师范大学, 2007.

[30] 孙 蓉, 孙秀玲, 宫雪亮, 等. 大汶河戴村坝径流序列一致性影响分析[J]. 水电能源科学, 2017, 35(09): 17-21.

基金资助

变化环境下山东省水循环变异与水资源请情势研究(SDSLKJ2024003)