因果分析-深度学习耦合的南水北调中线干渠水质预测方法

Causality analysis-deep learning based water quality analysis and prediction for the South-to-North Water Transfers Project

摘要

为实现南水北调工程的水质、水生态高精度预测，提出一种结合经验动态建模与深度学习的水质分析与预测方法，以南水北调中线工程藻类暴发风险较高的3个监测断面为研究对象，开展水质分析与预测研究。研究表明：研究区域内总磷与pH值、溶解氧之间存在双向因果关系，总氮与上述水质指标同样存在双向因果关联。进一步通过融合因果分析结果构建预测数据集可显著提升神经网络预测精度，纳什效率系数（E_NS）均超过0.9。与相关性分析相比，经验动态建模能够实现指标间的因果解析并揭示其内在关联规律。本研究构建方法可实现水质精准预测，为后续藻类生消过程分析与预测提供可靠技术支撑。

Abstract

The South-to-North Water Transfer Project (SNWTP) is a major strategic initiative designed to mitigate water scarcity in northern China. As a long-distance artificial channel, the SNWTP's ecological system has not yet reached a stable state, posing potential risks to water quality and aquatic ecology, particularly in sections susceptible to algal blooms. Effective monitoring and prediction of water quality parameters are crucial for maintaining water safety and sustainability. A hybrid method combining empirical dynamic modeling (EDM) with deep learning was proposed to analyze and forecast water quality at three vulnerable monitoring sections along the Middle Route of the SNWTP, which are at risk of algal outbreaks. Causal relationships among key water quality indicators were first examined using EDM. Subsequently, the identified causal variables were incorporated into the construction of predictive datasets, which were then used to train deep neural network models for water quality prediction. The application of EDM revealed bidirectional causal relationships between total phosphorus and both pH and dissolved oxygen, as well as as between total nitrogen and the same two parameters, within the studied sections. By integrating these causality-based indicators into the predictive framework, the deep learning models achieved high forecasting accuracy, with Nash-Sutcliffe efficiency coefficients exceeding 0.9. The proposed method demonstrated superior performance in capturing dynamic interactions among water quality variables compared with conventional correlation-based analyses. Unlike traditional correlation analysis, EDM effectively elucidated causal linkages among water quality indicators and provided deeper insight into their interdependent behavior. The combined EDM-deep learning approach offers a reliable and accurate tool for water quality prediction and supports subsequent analysis and early warning of algal bloom dynamics in large-scale water diversion projects.

关键词

因果分析 / 预测 / 经验动态建模 / 深度学习 / 水质

Key words

causal analysis / prediction / empirical dynamic modeling / deep learning / water quality

参考文献

[1] Bi Keyue, Wang Yongkang, Li Zhaoxin, et al. Traceability of gushing water in the Middle Route of the South-to-North Water Diversion (Beijing section) through the river area[J]. Journal of Environmental Management, 2024, 364: 121450. DOI: https://doi.org/10.1016/j.jenvman.2024.121450.

[2] Long Yan, Liu Yuxin, Zhao Tianhao, et al. Optimal allocation of water resources in the Middle Route of South-to-North Water Diversion Project based on multi-regional input-output model[J]. Journal of Hydrology, 2024, 637: 131381. DOI: https://doi.org/10.1016/j.jhydrol.2024.131381.

[3] Liu Mengkai, Feng Xiaobo, Wang Changde, et al. . Regulating a long-distance canal system with floating ice cover in the winter[J]. Journal of Cold Regions Engineering, 2015, 29(1): 04014009. DOI: 10.1061/(ASCE)CR.1943-5495.0000076.

[4] Wu Yongyan, Li Liqun, Liu Zihui, et al. Real-time control of the Middle Route of South-to-North Water Diversion Project[J]. Water, 2021, 13(1): 97. DOI: https://doi.org/10.3390/w13010097.

[5] Xia Rui, Zou Lei, Zhang Yuan, et al. Algal bloom prediction influenced by the Water Transfer Project in the Middle-lower Hanjiang River[J]. Ecological Modelling, 2022, 463: 109814. DOI: https://doi.org/10.1016/j.ecolmodel.2021.109814.

[6] Xu Zepeng, Liu Mengkai, Huang Minghai, et al. Application of real-time water temperature prediction system in winter for long-distance water diversion projects[J]. Journal of Hydroinformatics, 2024, 26(4): 915-933. DOI: https://doi.org/10.2166/hydro.2024.064.

[7] Zhu Jie, Lei Xiaohui, Quan Jin, et al. Algae growth distribution and key prevention and control positions for the Middle Route of the South-to-North Water Diversion Project[J]. Water, 2019, 11(9): 1851. DOI: https://doi.org/10.3390/w11091851.

[8] Wang Chao, Zhang Hong, Lei Pei, et al. Evidence on the causes of the rising levels of COD_Mn along the middle route of the South-to-North Diversion Project in China: The role of algal dissolved organic matter [J]. Journal of Environmental Sciences, 2022, 113: 281-290. DOI: https://doi.org/10.1016/j.jes.2021.06.003.

[9] Prasad D V V, Venkataramana L Y, Kumar P S, et al. Analysis and prediction of water quality using deep learning and auto deep learning techniques[J]. Science of the Total Environment, 2022, 821: 153311. DOI: https://doi.org/10.1016/j.scitotenv.2022.153311.

[10] Wan Hang, Xu Rui, Zhang Meng, et al. A novel model for water quality prediction caused by non-point sources pollution based on deep learning and feature extraction methods[J]. Journal of Hydrology, 2022, 612: 128081. DOI: https://doi.org/10.1016/j.jhydrol.2022.128081.

[11] Zheng Hang, Liu Yueyi, Wan Wenhua, et al. Large-scale prediction of stream water quality using an interpretable deep learning approach[J]. Journal of Environmental Management, 2023, 331: 117309. DOI: https://doi.org/10.1016/j.jenvman.2023.117309.

[12] Wang Yuefeng, Zhang Jiajun, Xie Lizi, et al. Water quality prediction for different time steps based on the ED-LSTM model[J]. South-to-North Water Transfers and Water Science & Technology, 2025, 23(6): 1522-1533. DOI: 10.13476/j.cnki.nsbdqk.2025.0150.

[13] Cheng Haomiao, Yang Fukang, Zhang Jian, et al. Research progress on river and lake water quality assessment based on machine learning[J]. Environmental Science, 2025, 46(6): 3349-3360. DOI: 10.13227/j.hjkx.202406089.

[14] Moran P A P. The statistical analysis of the Canadian lynx cycle[J]. Australian Journal of Zoology, 1953, 1(3): 291-298. DOI: 10.1071/ZO9530291.

[15] Ward A. Spurious correlations and causal inferences[J]. Erkenntnis, 2013, 78(3): 699-712. DOI: 10.1007/s10670-012-9411-6.

[16] Granger C W J. Investigating causal relations by econometric models and cross-spectral methods[J]. Econometrica: Journal of the Econometric Society, 1969: 424-438. DOI: https://doi.org/10.1017/CBO9780511753978.002.

[17] Huang Yu, Fu Zuntao, Franzke C L E. Detecting causality from time series in a machine learning framework[J]. Chaos: An Interdisciplinary Journal of Nonlinear Science, 2020, 30(6): 063116. DOI: 10.1063/5.0007670.

[18] Aguirregabiria V, Nevo A. Recent developments in empirical dynamic models of demand and competition in oligopoly markets[R]. Unpublished manuscript. July, 2010.

[19] Ye Hao, Beamish R J, Glaser S M, et al. Equation-free mechanistic ecosystem forecasting using empirical dynamic modeling[J]. Proceedings of the National Academy of Sciences, 2015, 112(13): E1569-E1576. DOI: https://doi.org/10.1073/pnas.1417063112.

[20] Altman N, Krzywinski M. Points of Significance: Association, correlation and causation[J]. Nature Methods, 2015, 12(10): 899-911. DOI: 10.1038/nmeth.3587.

[21] Vissio G, Lucarini V. A proof of concept for scale adaptive parametrizations: The case of the Lorenz'96 model[J]. Quarterly Journal of the Royal Meteorological Society, 2018, 144(710): 63-75. DOI: https://doi.org/10.1002/qj.3184.

[22] Wen Chunhui, Yang Jinhai. Complexity evolution of chaotic financial systems based on fractional calculus[J]. Chaos, Solitons & Fractals, 2019, 128: 242-251. DOI: https://doi.org/10.1016/j.chaos.2019.08.005.

[23] Cao Zhihao, Mu Shaomin, Xu Li, et al. Causal research on soil temperature and moisture content at different depths[J]. IEEE Access, 2021, 9: 39077-39088. DOI: 10.1109/access.2021.3064264.

[24] Flynn K F, Chapra S C, Suplee M W. Modeling the lateral variation of bottom-attached algae in rivers[J]. Ecological Modelling, 2013, 267: 11-25. DOI: https://doi.org/10.1016/j.ecolmodel.2013.07.011.

[25] Wang Yichong, Xu Chao, Lin Qianru, et al. Modeling of algal blooms: Advances, applications and prospects[J]. Ocean & Coastal Management, 2024, 255: 107250. DOI: https://doi.org/10.1016/j.ocecoaman.2024.107250.

[26] Chang Chunwei, Ushio M, Hsieh C. Empirical dynamic modeling for beginners[J]. Ecological Research, 2017, 32(6): 785-796. DOI: https://doi.org/10.1007/s11284-017-1469-9.

[27] Deyle E R, Sugihara G. Generalized theorems for nonlinear state space reconstruction[J]. PLoS ONE, 2011, 6(3): e18295. DOI: https://doi.org/10.1371/journal.pone.0018295.

[28] Tsay R S, Chen Rong. Nonlinear time series analysis[M]. Hoboken: John Wiley & Sons, 2018. DOI: https://doi.org/10.1017/CBO9780511755798.

[29] Sugihara G, May R M. Nonlinear forecasting as a way of distinguishing chaos from measurement error in time series[J]. Nature, 1990, 344(6268): 734-741. DOI: https://doi.org/10.1038/344734a0.

[30] Sugihara G. Nonlinear forecasting for the classification of natural time series[J]. Philosophical Transactions of the Royal Society of London. Series A: Physical and Engineering Sciences, 1994, 348(1688): 477-495. DOI: https://doi.org/10.1098/rsta.1994.0106.

[31] Le V Q M, Martinerie J, Adam C, et al. Nonlinear analyses of interictal EEG map the brain interdependences in human focal epilepsy[J]. Physica D: Nonlinear Phenomena, 1999, 127(3-4): 250-266. DOI: https://doi.org/10.1016/S0167-2789(98)00258-9.

[32] Van H G, Mosquera C, Napoles G. A review on the long short-term memory model[J]. Artificial Intelligence Review, 2020, 53(8): 5929-5955. DOI: https://doi.org/10.1007/s10462-020-09838-1.

[33] Graves A. Long short-term memory[J]. Supervised Sequence Labelling with Recurrent Neural Networks, 2012: 37-45. DOI: 10.1162/neco.1997.9.8.1735.

[34] Seiki T, Izawa H, Date E. Benthic nutrient remineralization and oxygen consumption in the coastal area of Hiroshima Bay[J]. Water Research, 1989, 23(2): 219-228. DOI: 10.1016/0043-1354(89)90046-8.

[35] Du L G, Rinklebe J, Vandecasteele B, et al. Trace metal behaviour in estuarine and riverine floodplain soils and sediments: A review[J]. Science of the Total Environment, 2009, 407(13): 3972-3985. DOI: https://doi.org/10.1016/j.scitotenv.2008.07.025.

[36] Pan Feng, Guo Zhanrong, Cai Yu, et al. Remobilization and hypoxia-dependent migration of phosphorus at the coastal sediment-water interface[J]. Journal of Hazardous Materials, 2021, 411: 125078. DOI: https://doi.org/10.1016/j.jhazmat.2021.125078.

[37] Tian Jing, Wang Gangsheng, Xiang Daifeng, et al. Causality analysis and prediction of riverine algal blooms by combining empirical dynamic modeling and machine learning techniques[J]. Water Resources Research, 2024, 60(5): e2023WR036334. DOI: https://doi.org/10.1029/2023WR036334.

[38] Ye Hao, Deyle E R, Gilarranz L J, et al. Distinguishing time-delayed causal interactions using convergent cross mapping[J]. Scientific Reports, 2015, 5(1): 14750. DOI: 10.1038/srep14750.

[39] Coufla D, Jakubik J, Jajcay N, et al. Detection of coupling delay: A problem not yet solved[J]. Chaos: An Interdisciplinary Journal of Nonlinear Science, 2017, 27(8): 083109. DOI: 10.1063/1.4997757.

[40] Ge Xinlei, Lin Aijing. Dynamic causality analysis using overlapped sliding windows based on the extended convergent cross-mapping[J]. Nonlinear Dynamics, 2021, 104(2): 1753-1765. DOI: https://doi.org/10.1007/s11071-021-06362-x.

[41] Yin Zhaokai, Liao Weihong, Wang Ruojia, et al. Rainfall-runoff modelling and forecasting based on long short-term memory (LSTM)[J]. South-to-North Water Transfers and Water Science & Technology, 2019, 17(6): 1-9. DOI: https://link.cnki.net/urlid/13.1430.tv.20250807.0852.002.

[42] Wang Yilin, Jin Yanguo, Chen Xiaonan, et al. Comparison between LSTM neural network and dimensional analysis method in discharge calculation of arc gates[J]. South-to-North Water Transfers and Water Science & Technology, 2022, 20(3): 590-599. DOI: 10.13476/j.cnki.nsbdqk.2022.0060.

[43] Gao Yiran, Lu Longbin, Sang Guoqing, et al. LSTM model for the real-time prediction of pre-station water level based on input factor analysis[J]. South-to-North Water Transfers and Water Science & Technology, 2025, 23(1): 157-165 DOI: 10.13476/j.cnki.nsbdqk.2025.0016.