深圳湾海水水质预测深度学习模型对比分析

doi:10.11978/2025134

摘要/Abstract

摘要：

深圳湾水体交换能力较弱, 且周边河流输入大量陆源污染物, 导致赤潮频发, 水质波动显著, 及时准确地预测深圳湾水质动态变化对该区域海洋灾害预警至关重要。传统水质预测模型或单一机器学习模型存在计算效率低、长期依赖捕捉能力不足等问题, 本文提出了一种基于变分模态分解的深度学习组合模型, 并对比分析该组合模型与基础深度学习模型在深圳湾水质预测中的表现。“分解—预测—重构”组合模型通过变分模态分解(variational mode decomposition, VMD)将水质和营养盐数据分解为5个平稳本征模态函数, 输入到时间卷积网络(temporal convolutional network, TCN)、长短期记忆网络(long short-term memory networks, LSTM)模型当中预测各个本征函数, 将预测结果叠加生成最终预测值。研究结果表明, (1)TCN-VMD和LSTM-VMD组合模型较基础TCN、LSTM模型预测效果显著提升, 纳什效率系数(Nash-Sutcliffe efficiency, NSE)最高改进率达到91.69%; (2)TCN-VMD相比LSTM-VMD纳什系数的表现更优, NSE最大改进率为95.56%, 对于水温(0.998)、pH(0.971)、盐度(0.988)、溶解氧(0.960)、叶绿素a(0.909)、氨氮(0.911)、磷酸盐(0.909)、硝酸盐(0.969)、亚硝酸盐(0.968), TCN-VMD均能准确捕捉真实值的波峰和波谷; (3)TCN-VMD为深圳湾水质预测提供了更精准高效的深度学习模型, 有效助力近岸海洋灾害预警。

关键词: 变分模态分解, 时间卷积网络, 长短期记忆网络, 水质预测, 深圳湾

Abstract:

Due to weak water exchange capacity and significant input of land-based pollutants from surrounding rivers, Shenzhen Bay experiences frequent red tide occurrences and marked seawater quality fluctuations. Timely and accurate prediction of dynamic changes in Shenzhen Bay's seawater quality is crucial for early warning of marine disasters in the region. Traditional water quality prediction models or single machine learning models face limitations such as low computational efficiency and insufficient capture of long-term dependencies. To address these challenges, this study proposes a hybrid deep learning model based on variational mode decomposition (VMD) and compares its performance with basic deep learning models in seawater quality prediction for Shenzhen Bay. The "decomposition-prediction-reconstruction" hybrid model first applies VMD to decompose seawater quality and nutrient data into five stable intrinsic mode functions (IMFs). These IMFs are then fed into temporal convolutional network (TCN) and long short-term memory (LSTM) models for individual prediction, with the final prediction value generated by summing all IMF predictions. The results demonstrate that: (1) The TCN-VMD and LSTM-VMD hybrid models significantly outperform the basic TCN and LSTM models, with the highest improvement in Nash-Sutcliffe efficiency coefficient (NSE) reaching 91.69%; (2) TCN-VMD performs better than LSTM-VMD in terms of NSE, achieving a maximum improvement rate of 95.56%. For water temperature (0.998), pH (0.971), salinity (0.988), dissolved oxygen (0.960), chlorophyll-a (0.909), ammonia nitrogen (0.911), phosphate (0.909), nitrate (0.969), and nitrite (0.968), TCN-VMD accurately captures the peaks and troughs of the true values; (3) TCN-VMD provides a more accurate and efficient deep learning framework for seawater quality prediction in Shenzhen Bay, effectively supporting early warning of coastal marine disasters.

Key words: VMD, TCN, LSTM, seawater quality prediction, Shenzhen Bay

中图分类号:

X834

王凯敏, 陈瑶, 陈俊嘉, 陈正茜. 深圳湾海水水质预测深度学习模型对比分析[J]. 热带海洋学报, 2026, 45(3): 188-201.

WANG Kaimin, CHEN Yao, CHEN Junjia, CHEN Zhengxi. Comparative analysis of deep learning models for seawater quality prediction in Shenzhen Bay[J]. Journal of Tropical Oceanography, 2026, 45(3): 188-201.

图/表 16

图1

表1

表2

图2

表3

图3

图4

图5

图6

图7

图8

图9

图10

图11

图12

图13

参考文献 26

[1]	曹文治, 苏雅, 曾阳艳, 等, 2023. 基于EEMD-LSTM-SVR的水质预测模型[J]. 系统工程, 41(4): 1-12.
	CAO WENZHI, SU YA, ZENG YANGYAN, et al, 2023. Water quality prediction model based on EEMD-LSTM-SVR[J]. Systems Engineering, 41(4): 1-12 (in Chinese with English abstract).
[2]	陈亚松, 刘家雯, 赵云鹏, 等, 2025. 基于机器学习的人工湿地出水水质预测与影响因素[J]. 中国环境科学, 45(6): 3161-3170.
	CHEN YASONG, LIU JIAWEN, ZHAO YUNPENG, et al, 2025. Prediction of effluent water quality and analysis of influencing factors in constructed wetlands based on machine learning[J]. China Environmental Science, 45(6): 3161-3170 (in Chinese with English abstract).
[3]	陈一帆, 李泽, 周成龙, 等, 2024. 基于两层分解和极限学习机的河流水质预测[J]. 计算机仿真, 41(6): 536-543+582.
	CHEN YIFAN, LI ZE, ZHOU CHENGLONG, et al, 2024. Prediction of river water quality based on two-layer decomposition method and extreme learning machine[J]. Computer Simulation, 41(6): 536-543+582 (in Chinese with English abstract).
[4]	侯德, 2024. 两种机器学习算法在水质预测中的应用[J]. 环境污染与防治, 46(11): 1596-1600+1607.
	HOU DE, 2024. Application of two machine learning algorithms in water quality prediction[J]. Environmental Pollution & Control, 46(11): 1596-1600+1607 (in Chinese with English abstract).
[5]	桓清柳, 庞仁松, 周秋伶, 等, 2016. 深圳近岸海域氮、磷营养盐变化趋势及其与赤潮发生的关系[J]. 海洋环境科学, 35(6): 908-914.
	HUAN QINGLIU, PANG RENSONG, ZHOU QIULING, et al, 2016. Variation trends of nitrogen and phosphorus and the relationship with HABs in Shenzhen coastal waters[J]. Marine Environmental Science, 35(6): 908-914 (in Chinese with English abstract).
[6]	李明, 赵良伟, 蒋一波, 等, 2024. 基于分解-预测-灰狼优化集成模型的水质预测研究[J]. 中国农村水利水电(11): 95-102.
	LI MING, ZHAO LIANGWEI, JIANG YIBO, et al, 2024. Water quality level prediction based on the decomposition-prediction-gray wolf optimization ensemble model[J]. China Rural Water and Hydropower, (11): 95-102 (in Chinese with English abstract).
[7]	梅丹, 张恒, 2024. 基于变分模态分解和深度学习算法的污水出水水质预测[J]. 长江科学院院报, 42(9): 67-82.
	MEI DAN, ZHANG HENG, 2024. Prediction of wastewater effluent water quality based on variational mode decomposition and deep learning algorithm[J]. Journal of Changjiang River Scientific Research Institute: 42(9): 67-82 (in Chinese with English abstract).
[8]	孙金水, WAI ONYX WING-HONG, 王伟, 等, 2010. 深圳湾海域氮磷营养盐变化及富营养化特征[J]. 北京大学学报(自然科学版), 46(6): 960-964.
	SUN JINSHUI, WAI ONYX WING-HONG, WANG WEI, et al, 2010. Characteristics of eutrophication and variation trend of nitrogen and phosphorus nutrient in Shenzhen Bay[J]. Acta Scientiarum Naturalium Universitatis Pekinensis, 46(6): 960-964 (in Chinese with English abstract).
[9]	苏金洙, 邹嘉澍, 苏玉萍, 等, 2022. 福建平潭近海赤潮预警模型研究[J]. 热带海洋学报, 41(4): 172-180. doi: 10.11978/2021156
	SU JINZHU, ZOU JIASHU, SU YUPING, et al, 2022. Study on the early warning model of red tide in the offshore area of Pingtan, Fujian Province[J]. Journal of Tropical Oceanography, 41(4): 172-180 (in Chinese). doi: 10.11978/2021156
[10]	唐毅, 徐全, 杜彬, 等, 2024. 基于SARIMA-VMD-LSSVM的水产养殖溶解氧质量浓度预测[J]. 江苏农业学报, 40(8): 1473-1482.
	TANG YI, XU QUAN, DU BIN, et al, 2024. Prediction of dissolved oxygen mass concentration in aquaculture based on SARIMA-VMD-LSSVM[J]. Jiangsu Journal of Agricultural Sciences, 40(8): 1473-1482 (in Chinese with English abstract).
[11]	翁少佳, 蔡锦海, 庞运禧, 等, 2024. 卷积神经网络在近岸表层海温预报中的应用[J]. 热带海洋学报, 43(1): 40-47.
	WENG SHAOJIA, CAI JINHAI, PANG YUNXI, et al, 2024. Application of convolutional neural network to sea surface temperature prediction in the coastal waters[J]. Journal of Tropical Oceanography, 43(1): 40-47 (in Chinese with English abstract).
[12]	项新建, 许宏辉, 谢建立, 等, 2024a. 基于VMD-TCN-GRU模型的水质预测研究[J]. 人民黄河, 46(3): 92-97.
	XIANG XINJIAN, XU HONGHUI, XIE JIANLI, et al, 2024. Water quality prediction based on VMD-TCN-GRU model[J]. Yellow River, 46(3): 92-97 (in Chinese with English abstract).
[13]	项新建, 张颖超, 许宏辉, 等, 2024b. 基于CEEMDAN-VMD-TCN-lightGBM模型的水质预测研究[J]. 中国农村水利水电, (3): 86-95.
	XIANG XINJIAN, ZHANG YINGCHAO, XU HONGHUI, et al, 2024. Research on water quality prediction based on CEEMDAN-VMD-TCN-light GBM model[J]. China Rural Water and Hydropower, (3): 86-95 (in Chinese with English abstract).
[14]	肖明君, 朱逸纯, 高雯媛, 等, 2024. 基于不同人工神经网络的水质预测方法对比[J]. 环境科学, 45(10): 5761-5767.
	XIAO MINGJUN, ZHU YICHUN, GAO WENYUAN, et al, 2024. Comparative study of water quality prediction methods based on different artificial neural network[J]. Environmental Science, 45(10): 5761-5767 (in Chinese with English abstract). doi: 10.1021/es1041755
[15]	肖裕锋, 张代青, 2025. 可解释性融合神经网络在巢湖总磷浓度预测中的应用[J]. 中国农村水利水电, (7): 44-51.
	XIAO YUFENG, ZHANG DAIQING, 2025. Application of interpretable fusion neural network in the prediction of total phosphorus concentration in Chaohu Lake[J]. China Rural Water and Hydropower, (7): 44-51 (in Chinese with English abstract).
[16]	谢小良, 吴琳琳, 2024. 基于DA-CNN-BiLSTM的河流溶解氧浓度预测[J]. 人民黄河, 46(7): 92-97+111.
	XIE XIAOLIANG, WU LININ, 2024. River dissolved oxygen prediction based on DA-CNN-BiLSTM[J]. Yellow River, 46(7): 92-97+111 (in Chinese with English abstract).
[17]	谢雨茜, 李路, 朱明, 等, 2022. 基于EMD与K-means的ILSTM模型在池塘溶解氧预测中的应用[J]. 华中农业大学学报, 41(3): 200-210.
	XIE YUXI, LI LU, ZHU MING, et al, 2022. Application of ILSTM model based on EMD and K-means in prediction of dissolved oxygen in pond[J]. Journal of Huazhong Agricultural University, 41(3): 200-210 (in Chinese).
[18]	熊剑智, 熊睿, 鲁海燕, 等, 2024. 基于机器学习的深圳湾水质预报[J]. 人民珠江, 45(7): 10-18.
	XIONG JIANZHI, XIONG RUI, LU HAIYAN, et al, 2024. Machine learning-based water quality forecasting for Shenzhen Bay[J]. Pearl River, 45(7): 10-18 (in Chinese with English abstract).
[19]	张景平, 黄小平, 江志坚, 等, 2010. 珠江口海域污染的水质综合污染指数和生物多样性指数评价[J]. 热带海洋学报, 29(1): 69-76.
	ZHANG JINGPING, HUANG XIAOPING, JIANG ZHIJIAN, et al, 2010. Assessment of the Pearl River Estuary pollution by water comprehensive pollution index and biodiversity index[J]. Journal of Tropical Oceanography, 29(1): 69-76 (in Chinese with English abstract).
[20]	张海霞, 李瑞, 王霞, 等, 2025. 基于VMD和GA-BiLSTM组合模型的河流水质预测[J]. 环境工程技术学报, 15(4): 1181-1188.
	ZHANG HAIXIA, LI RUI, WANG XIA, et al, 2025. River water quality prediction based on the combined model of VMD and GA-BiLSTM[J]. Journal of Environmental Engineering Technology, 15(4): 1181-1188 (in Chinese with English abstract).
[21]	张颖, 李梅, 高倩倩, 等, 2016. 基于ELMR-SVMR的海水水质预警模型研究[J]. 大连理工大学学报, 56(2): 185-192.
	ZHANG YING, LI MEI, GAO QIANQIAN, et al, 2016. Research on forewarning model of seawater quality based on ELMR-SVMR[J]. Journal of Dalian University of Technology, 56(2): 185-192 (in Chinese with English abstract).
[22]	张军晓, 李绪录, 周毅频, 等, 2013. 2000—2010年深圳湾及其邻近海域溶解无机氮的时空分布[J]. 生态环境学报, 22(3): 475-480.
	ZHANG JUNXIAO, LI XULU, ZHOU YIPIN, et al, 2013. Temporal and spatial distributions of dissolved inorganic nitrogen in the Shenzhen Bay and its adjacent zone from 2000 to 2010[J]. Ecology and Environmental Sciences, 22(3): 475-480 (in Chinese with English abstract).
[23]	张思萱, 康燕, 宋金玲, 等, 2024. 基于OVMD-TCN-AR的水质预测模型[J]. 环境科学导刊, 43(5): 61-66.
	ZHANG SIXUAN, KANG YAN, SONG JINLING, et al, 2024. Water quality prediction model based on OVMD-TCN-AR[J]. Environmental Science Survey, 43(5): 61-66 (in Chinese with English abstract).
[24]	张静, 张瑜斌, 周凯, 等, 2010. 深圳湾海域营养盐的时空分布及潜在性富营养化程度评价[J]. 生态环境学报, 19(2): 253-261. doi: 10.16258/j.cnki.1674-5906(2010)02-0253-09
	ZHANG JING, ZHANG YUBIN, ZHOU KAI, et al, 2010. Evaluation on temporal and spatial distribution of nutrients and potential eutrophication in Shenzhen Bay[J]. Ecology and Environment, 19(2): 253-261 (in Chinese with English abstract).
[25]	张瑜斌, 章洁香, 张才学, 等, 2009. 赤潮多发区深圳湾叶绿素a的时空分布及其影响因素[J]. 生态环境学报, 18(5): 1638-1643. doi: 10.16258/j.cnki.1674-5906(2009)05-1638-06
	ZHANG YUBIN, ZHANG JIEXIANG, ZHANG CAIXUE, et al, 2009. The spatial and temporal distribution of chlorophyll a and the influencing factors in Shenzhen Bay with frequent occurrence of red tide[J]. Ecology and Environmental Sciences, 18(5): 1638-1643 (in Chinese with English abstract).
[26]	AMIREDDY R, DILEEP P, 2024. A comparative study on water quality prediction using machine learning and deep learning techniques. 2024 Third International Conference on Distributed Computing and Electrical Circuits and Electronics (ICDCECE)[C]. Ballari, India: IEEE: 1-5.

	PyPots	SVR	线性插值	样条插值
水质参数误差	1.00	1.02	0.25	0.34
营养盐参数误差	0.57	1.08	0.57	1.05

参数	TCN/TCN-VMD	LSTM/LSTM-VMD
卷积核大小	3	——
扩张因子	[1, 2, 4, 8]	——
隐藏层大小	——	128
隐藏层层数	4	3
批次大小	64	64
训练次数	100	100
水质输入特征	8	8
水质输出特征	5	5
营养盐输入特征	4	4
营养盐输出特征	4	4
水质滑动窗口大小	288	288
营养盐滑动窗口大小	100	100
丢弃率	0.6	0.6
学习率	0.0005	0.0005
预测步长	1	1
IMF个数	5	5

	TCN	TCN-VMD	LSTM	LSTM-VMD
水温	0.984	0.998	0.949	0.955
pH	0.927	0.971	0.864	0.842
盐度	0.959	0.988	0.908	0.961
溶解氧	0.925	0.960	0.795	0.912
叶绿素a	0.798	0.909	0.805	0.842
氨氮	0.596	0.911	-0.159	0.894
磷酸盐	0.574	0.909	0.323	0.691
硝酸盐	0.816	0.969	0.627	0.969
亚硝酸盐	0.850	0.968	0.214	0.892