信号分解与深度学习集成策略在水质预测中的应用

郑芷延; 关翔鸿; 邱光磊; 吴海珍; 韦朝海

doi:10.13205/j.hjgc.202610008

信号分解与深度学习集成策略在水质预测中的应用

doi: 10.13205/j.hjgc.202610008

1. 华南理工大学环境与能源学院, 广州 510006;
2. 华南理工大学生物科学与工程学院, 广州 510006

基金项目:

国家重点研发计划（2023YFC3207005）

详细信息

作者简介:
郑芷延（2001—），女，硕士研究生，主要研究方向为数据模型构建模拟。202320147245@mail.scut.edu.cn

通讯作者:
韦朝海（1962—），男，教授，主要研究方向为水污染控制理论与技术。cechwei@scut.edu.cn

计量
- 文章访问数: 4
- HTML全文浏览量: 0
- PDF下载量: 1
- 被引次数: 0
出版历程
- 收稿日期: 2026-03-26
- 网络出版日期: 2026-11-10

Application of signal decomposition and deep learning integration strategies in water quality prediction

1. School of Environment and Energy, South China University of Technology, Guangzhou 510006, China;
2. School of Biomedical and Engineering, South China University of Technology, Guangzhou 510006, China

摘要

摘要: 精准的水质预测是推动水环境管理模式由事后响应转向事前防控转变的关键环节。针对水质时间序列的非线性、非平稳与多尺度的数据特性及其准确预测难题，提出了融合信号分解与混合模型的水质预测策略。以我国多个国控监测断面2021年1月—2024年12月的水质监测数据为基础，构建了通过贝叶斯优化的CEEMDAN-TCN-LSTM模型，其仅依赖常规水质参数，在多指标、多站点和多步长3个维度上展现出良好的鲁棒性与泛化性能，平均决定系数（R²）分别为0.93、0.96和0.96。采用的完全自适应噪声集合经验模态分解（CEEMDAN）能够有效提取水质序列的多尺度信息，使混合神经网络（TCN-LSTM）模型预测误差降低68%；针对信号分解导致输入变量激增的问题，引入样本熵与重组规则，在精度损失可接受范围内使计算负荷削减50%以上。夏普利加性解释（SHAP）结果显示：溶解氧时间序列中各分量的特征变量重要性排序存在明显差异，同一指标的贡献模式也随尺度变化，强调了多尺度信息在提升水质预测模型对复杂波动响应能力方面的必要性。研究结果立足水质时间序列的多尺度特征，充分利用现有河流断面监测系统水质数据资源，为推动智能化水环境管理提供了技术支持。
- 水质预测 /
- 信号分解 /
- LSTM-TCN模型 /
- 贝叶斯优化 /
- SHAP可解释性分析
Abstract: Accurate water quality prediction serves as a key step in facilitating the transformation of water environment management models from post-event response to proactive prevention. To address the nonlinear， non-stationary， and multi-scale characteristics of water quality time series and the challenges of accurate prediction， this study proposed a novel water quality prediction strategy that integrates signal decomposition with deep learning models. Based on water quality monitoring data collected from multiple state-controlled monitoring sections from January 2021 to December 2024， a CEEMDAN-TCN-LSTM water quality prediction model via Bayesian optimization was eatablished relying solely on routine water quality parameters. The model demonstrated excellent robustness and generalization performance across three dimensions， with average coefficients of determination （R²） of 0.93 for multi-indicator predictions， 0.96 for multi-site predictions， and 0.96 for multi-step predictions. The employed complete ensemble empirical mode decomposition with adaptive noise （CEEMDAN） effectively extracted multi-scale information from the water quality sequences， reducing the prediction error of the hybrid neural network （TCN-LSTM） model by 68%. To address the surge in input variables caused by signal decomposition， this study introduced Sample Entropy （SampEn） alongside adaptive recombination rules； this approach successfully reduced the computational load by over 50% while keeping accuracy loss within a strictly defined acceptable range. Furthermore， Bayesian optimization （BO） improved the prediction accuracy and operational flexibility of the model. By breaking the black-box limitation of data-driven models， Shapley additive explanations （SHAP） analysis revealed significant differences in the importance ranking of feature variables across different components of the dissolved oxygen （DO） time series， and the contribution pattern of individual indicators also varied with scale， underscoring the necessity of incorporating multi-scale information to enhance the responsiveness of water quality prediction models to complex fluctuations. Grounded in the multi-scale characteristics of water quality time series， this study enriches the methodology of water quality prediction modeling， makes full use of existing water quality data resources from river section monitoring systems， and provides technical support for advancing intelligent water environment management.
- water quality prediction /
- signal decomposition /
- LSTM-TCN model /
- Bayesian optimization /
- SHAP interpretability analysis

HTML全文

参考文献(30)

[1]	LI G,ZHANG Q,DU H,et al. Artificial intelligence prediction of water quality of complex urban river networks[J]. Water Resources Management,2025,39（12）:6421-6435.
[2]	ZHI W,APPLING A P,GOLDEN H E,et al. Deep learning for water quality[J]. Nature Water,2024,2（3）:228-241.
[3]	ZENG H B,LONG Q,GAO J H,et al. Application of machine learning in water quality prediction and analysis in river segments[J]. Environmental Engineering,2026,44（5）:50-60. 曾鸿滨,龙琦,高景恒,等. 机器学习在河流断面水质预测分析中的应用[J]. 环境工程,2026,44（5）:50-60.
[4]	MONTES C,KAPELAN Z,SALDARRIAGA J. Predicting non-deposition sediment transport in sewer pipes using Random forest[J]. Water Research,2021,189:116639.
[5]	LI R,FENG K,AN T,et al. Enhanced insights into effluent prediction in wastewater treatment plants:comprehensive deep learning model explanation based on SHAP[J]. ACS ES &T Water,2024,4（4）:1904-1915.
[6]	REN Y,WANG S,XIA B. Deep learning coupled model based on TCN-LSTM for particulate matter concentration prediction[J]. Atmospheric Pollution Research,2023,14（4）:101703.
[7]	BIEROZA M,ACHARYA S,BENISCH J,et al. Advances in catchment science,hydrochemistry,and aquatic ecology enabled by high-frequency water quality measurements[J]. Environ Sci Technol,2023,57（12）:4701-4719.
[8]	ZHANG Y,LI C,JIANG Y,et al. Accurate prediction of water quality in urban drainage network with integrated EMD-LSTM model[J]. Journal of Cleaner Production,2022,354:131724.
[9]	HUAN J,FAN Y,XU X,et al. Deep learning model based on coupled SWAT and interpretable methods for water quality prediction under the influence of non-point source pollution[J]. Computers and Electronics in Agriculture,2025,231:109985.
[10]	ARIYAEI A,KHEYRURI Y,DOROUDI S,et al. A hybrid EMD-DFA-LSTM-MOOTLBO model for accurate water quality index prediction[J]. Journal of Water Process Engineering,2025,77:108558.
[11]	SABZIPOUR B,ARSENAULT R,TROIN M,et al. Comparing a long short-term memory（LSTM）neural network with a physically-based hydrological model for streamflow forecasting over a Canadian catchment[J]. Journal of Hydrology,2023,627:130380.
[12]	TORRES M E,COLOMINAS M A,SCHLOTTHAUER G,et al. A complete ensemble empirical mode decomposition with adaptive noise[C]// Proceedings of the ICASSP,IEEE International Conference on Acoustics,Speech and Signal Processing,F,2011.
[13]	HENG L,HAO C,NAN L C. Load forecasting method based on CEEMDAN and TCN-LSTM[J]. PLOS ONE,2024,19（7）:e0300496.
[14]	LIN Y,KOPRINSKA I,RANA M. Temporal convolutional attention neural networks for time series forecasting[C]// Proceedings of the 2021 International Joint Conference on Neural Networks（IJCNN）,2021.
[15]	ZHANG Y T. Research on deep learning hybrid model for water quality prediction of urban sewage networks[D]. Harbin:Harbin Institute of Technology,2024. 张怡拓. 用于城市污水管网水质预测的深度学习混合模型研究[D]. 哈尔滨:哈尔滨工业大学,2024.
[16]	MENG G L,CONG Z L,SONG B,et al. A survey of Bayesian network structure learning[J]. Journal of Beijing University of Aeronautics and Astronautics,2025,51（9）:2829-2849. 孟光磊,丛泽林,宋彬,等. 贝叶斯网络结构学习综述[J]. 北京航空航天大学学报,2025,51（9）:2829-2849.
[17]	ARASHPOUR M. AI explainability framework for environmental management research[J]. Journal of Environmental Management,2023,342:118149.
[18]	HUANG H,GUAN X H,CHENG X Q,et al. New environmental risks caused by the imbalance of element proportion after the peaking of water pollution:A case study of Guangdong coastal area[J]. Journal of Environmental Sciences,2024,44（5）:138-148. 黄华,关翔鸿,成晓倩,等. 水污染达峰后元素比例失衡引发的新环境风险:以广东沿海为例[J]. 环境科学学报,2024,44（5）:138-148.
[19]	JIANG Y,LI C,SUN L,et al. A deep learning algorithm for multi-source data fusion to predict water quality of urban sewer networks[J]. Journal of Cleaner Production,2021,318:128533.
[20]	DING H N. Spatial-temporal characteristics analysis and prediction ofriver water quality in Pearl River（Guangzhou Section）[D]. Guangzhou:South China University of Technology,2023. 丁号楠. 珠江（广州段）流域河流水质时空特征分析和预测研究[D]. 广州:华南理工大学,2023.
[21]	BAND S S,QASEM S N,MANSOR Z,et al. Explainable deep learning hybrid models for enhanced prediction of river dissolved oxygen concentration[J]. Journal of Water Process Engineering,2026,81:109261.
[22]	ZHANG Y,WANG J,LI C,et al. Attention-based deep learning models for predicting anomalous shock of wastewater treatment plants[J]. Water Research,2025,275:123192.
[23]	NOORI N,KALIN L,ISIK S. Water quality prediction using SWAT-ANN coupled approach[J]. Journal of Hydrology,2020,590:125220.
[24]	JIANG J,GAO Y,CHEN J,et al. Effects of multi-step-ahead prediction strategies on LSTM-based runoff prediction[J]. Water Resources Management,2025,39（15）:8103-8116.
[25]	HOZ J L M D L,BAPPY M M,ISLAM M D S,et al. Interpretable forecasting of dissolved oxygen leveraging foundation model for proactive aeration in rural wastewater treatment systems[J]. Water Research,2026,289:124931.
[26]	ZOU X,WANG S,MAI W,et al. Explainable aeration prediction using deep learning with interpretability analysis[J]. Journal of Water Process Engineering,2025,71:107218.
[27]	CUI J,WANG Y,DING S,et al. High-resolution diurnal variation mechanism of oxygen and acid environments at the water-sediment interface during cyanobacterial decomposition[J]. Journal of Cleaner Production,2024,435:140605.
[28]	ZHANG W,RONG N,JIN X,et al. Dissolved oxygen variation in the North China Plain river network region over 2011-2020 and the influencing factors[J]. Chemosphere,2022,287（Pt 3）:132354.
[29]	TIAN Y X,GUAN X H,ZHENG Z Y,et al. Analysis of drivers and ecological response mechanisms of river eutrophication in China from an elemental balancing perspective[J]. Environmental Engineering,2025,43（12）:1-12. 田雨馨,关翔鸿,郑芷延,等. 元素平衡视角下我国河流富营养化驱动及生态响应机制分析[J]. 环境工程,2025,43（12）:1-12.
[30]	ZHANG Y,LI C,DUAN H,et al. Deep learning based data-driven model for detecting time-delay water quality indicators of wastewater treatment plant influent[J]. Chemical Engineering Journal,2023,467:143483.