基于动态滑动窗口BP神经网络的水质时间序列预测

张梦迪; 徐庆; 刘振鸿; 马春燕; 高品

doi:10.12153/j.issn.1674-991X.20210194

基于动态滑动窗口BP神经网络的水质时间序列预测

doi: 10.12153/j.issn.1674-991X.20210194

张梦迪^1,,
徐庆²,
刘振鸿¹,
马春燕¹,
高品^1, ,

1.
东华大学环境科学与工程学院
2.
上海市环境监测中心

基金项目: 上海市生态环境局科研项目（沪环科〔2020〕第51号）

详细信息

作者简介:
张梦迪(1996—)，女，硕士，研究方向为水污染控制技术，1393741826@qq.com

通讯作者:
高品（1982—），男，教授，博士，研究方向为水污染控制技术，pingao@dhu.edu.cn

中图分类号: X824
计量
- 文章访问数: 723
- HTML全文浏览量: 254
- PDF下载量: 100
- 被引次数: 0
出版历程
- 收稿日期: 2021-05-21
- 录用日期: 2021-09-30
- 网络出版日期: 2022-06-07

Prediction of water quality time series based on the dynamic sliding window BP neural network model

1.
College of Environmental Science and Engineering, Donghua University
2.
Shanghai Environmental Monitoring Center

摘要

摘要:
为提高BP神经网络（BPNN）模型对具有时间序列特征水质的预测精准度，采用主成分分析法对原始样本数据进行特征提取和降维，选取溶解性有机碳（DOC）浓度、总氮（TN）浓度和浊度作为水质预测指标，构建了具有3层网络结构的BPNN模型进行预测，并分析其预测性能。结果表明：DOC浓度、TN浓度和浊度的最佳训练集尺寸分别为60、60和90 d，最佳BPNN拓扑结构分别为9-12-1、8-6-1和7-13-1，经优化后的BPNN模型对DOC浓度、TN浓度和浊度的变化趋势整体预测效果较好；相比之下，BPNN模型对水中DOC浓度的预测效果显著优于TN浓度和浊度，其均方根误差（RMSE）、平均绝对百分比误差（MAPE）和相关系数（R）分别为0.040、0.66%和0.867。该模型对具有非线性特征的地表水水质预测具有较好的适用性，预测精度较高。
- BP神经网络 /
- 动态滑动窗口 /
- 水质时间序列 /
- 主成分分析法
Abstract:
In order to improve the prediction precision of water quality having time series properties by BP neural network (BPNN), principal component analysis (PCA) was used for characteristic extraction and dimension reduction of the original data. Concentrations of dissolved organic compound (DOC) and total nitrogen (TN), and turbidity were selected as the water quality prediction indices, a three-layer BPNN model was established for prediction, and the prediction performance was analyzed. The results showed that the optimal training-set sizes of concentrations of DOC and TN, and turbidity were 60, 60, and 90 days, while the best BPNN topological structures were 9-12-1, 8-6-1 and 7-13-1, respectively. The optimized BPNN model exhibited favorable prediction performance on the variation trends of concentrations of DOC and TN, and turbidity. In contrast, the prediction performance of DOC by BPNN model was significantly better than that of TN and turbidity, with RMSE, MAPE, and R values of 0.040, 0.66% and 0.867, respectively. This model had a good applicability and precision for prediction of surface water qualities having non-linear characteristics.
- BP neural network /
- dynamic sliding window /
- water quality time series /
- principal component analysis

HTML全文

图 1 滑动窗口示意

Figure 1. Schematic diagram of the sliding window

下载: 全尺寸图片幻灯片

图 2 BPNN拓扑结构基本框架

Figure 2. Basic framework of BPNN topology

下载: 全尺寸图片幻灯片

图 3 不同训练集尺寸的RMSE、R和MAPE对比

Figure 3. Comparison of RMSE, R and MAPE values of different training-set sizes

下载: 全尺寸图片幻灯片

图 4 不同滑动窗口大小的RMSE、R和MAPE对比

Figure 4. Comparison of RMSE, R and MAPE values with different sliding window sizes

下载: 全尺寸图片幻灯片

图 5 不同隐含层单元数的RMSE、R和MAPE对比

Figure 5. Comparison of RMSE, R and MAPE values of different hidden layer units

下载: 全尺寸图片幻灯片

图 6 优化BPNN模型对DOC浓度、TN浓度和浊度的预测效果

Figure 6. Prediction performance of optimized BPNN model towards DOC concentration, TN concentration, and turbidity

下载: 全尺寸图片幻灯片

表 1 主成分分析结果

Table 1. Results of principal component analysis

成分	特征值	贡献率/％
1	2.924	26.6
2	2.419	48.6
3	2.094	67.6
4	0.884	75.7
5	0.728	82.3

下载: 导出CSV

表 2 主成分系数矩阵

Table 2. Coefficient matrix of the principal components

主成分	1	2	3	4	5
pH	−0.626	0.171	−0.253	0.212	0.245
浊度	−0.392	0.813	0.207	−0.010	0.017
DO	0.441	0.463	−0.682	0.001	−0.066
水温	−0.766	−0.321	0.470	0.146	0.068
电导率	0.573	0.209	0.082	0.657	0.262
TN浓度	0.809	−0.098	0.252	0.180	−0.097
NH₃-N浓度	0.573	−0.039	0.535	−0.129	−0.338
TP浓度	0.077	0.613	0.552	−0.293	0.143
水中油浓度	0.421	−0.367	−0.339	−0.481	0.516
DOC浓度	0.146	0.890	0.119	−0.089	0.128
ORP	−0.286	0.230	−0.714	−0.104	−0.405

下载: 导出CSV

表 4 DOC浓度、TN浓度和浊度的最优模型参数

Table 4. Optimized model parameters of DOC concentration, TN concentration, and turbidity

水质指标	训练集尺寸/d	滑动窗口大小	隐含层单元数	模型评价指标
水质指标	训练集尺寸/d	滑动窗口大小	隐含层单元数	RMSE	R	MAPE/％
DOC浓度	60	9	12	0.051	0.876	0.68
TN浓度	60	8	6	0.591	0.858	10.45
浊度	90	7	13	14.737	0.749	23.60

下载: 导出CSV

参考文献(21)

[1]	史斌, 姜继平, 王鹏.基于高频在线水质数据异常的突发污染预警[J]. 中国环境科学,2017,37(11):4394-4400. doi: 10.3969/j.issn.1000-6923.2017.11.046 SHI B, JIANG J P, WANG P. Early warning of water pollution incidents based on abnormal change of water quality data from high frequency online monitoring[J]. China Environmental Science,2017,37(11):4394-4400. doi: 10.3969/j.issn.1000-6923.2017.11.046
[2]	MAHAJAN A U, CHALAPATIRAO C V, GADKARI S K. Mathematical modeling: a tool for coastal water quality management[J]. Water Science and Technology,1999,40(2):151-157. doi: 10.2166/wst.1999.0110
[3]	Fang X B, Zhang J Y, Chen Y X. QUAL2K model used in the water quality assessment of Qiantang River, China[J]. Water Environment Research,2008,80(11):2125-2133. doi: 10.2175/106143008X304794
[4]	沈煜恒, 戴小鹏, 湛誉, 等.基于灰色模型的水质监测预警系统设计[J]. 农业工程,2019,9(8):31-34. doi: 10.3969/j.issn.2095-1795.2019.08.011 SHEN Y H, DAI X P, ZHAN Y, et al. Design of water quality monitoring and early warning system based on grey model[J]. Agricultural Engineering,2019,9(8):31-34. doi: 10.3969/j.issn.2095-1795.2019.08.011
[5]	李如忠.水质预测理论模式研究进展与趋势分析[J]. 合肥工业大学学报(自然科学版),2006,29(1):26-30.
[6]	黄俊, 周申范, 唐婉莹.TNT生化降解时间序列的人工神经网络预报模型[J]. 环境科学研究,2000,13(2):3-5. doi: 10.3321/j.issn:1001-6929.2000.02.002 HUANG J, ZHOU S F, TANG W Y. Artificial neutral network predicting model of TNT biodegradation time series[J]. Research of Environmental Sciences,2000,13(2):3-5. doi: 10.3321/j.issn:1001-6929.2000.02.002
[7]	李春华, 胡文, 叶春, 等.基于BP神经网络预测地表水净化装置总氮的去除效果[J]. 环境工程技术学报,2018,8(6):651-655. doi: 10.3969/j.issn.1674-991X.2018.06.086 LI C H, HU W, YE C. Study on prediction of total nitrogen removal of a surface water purification device based on BP neural network[J]. Journal of Environmental Engineering Technology,2018,8(6):651-655. doi: 10.3969/j.issn.1674-991X.2018.06.086
[8]	林佳敏, 陈金良, 林晶晶, 等.BP神经网络和ARIMA模型对污水处理厂出水总氮浓度的模拟预测[J]. 环境工程技术学报,2019,9(5):573-578. doi: 10.12153/j.issn.1674-991X.2019.03.261 LIN J M, CHEN J L, LIN J J, et al. The simulation and prediction of TN in wastewater treatment effluent using BP neural network and ARIMA model[J]. Journal of Environmental Engineering Technology,2019,9(5):573-578. doi: 10.12153/j.issn.1674-991X.2019.03.261
[9]	左朝晖, 李绍康, 杨津津, 等.基于GA-BP神经网络的页岩气开发区域水资源承载力研究[J]. 环境工程技术学报,2021,11(1):194-201. doi: 10.12153/j.issn.1674-991X.20200081 ZUO Z H, LI S K, YANG J L, et al. Research on water resources carrying capacity of shale gas development area based on GA-BP neural network[J]. Journal of Environmental Engineering Technology,2021,11(1):194-201. doi: 10.12153/j.issn.1674-991X.20200081
[10]	刘宇同. 一种基于人工神经网络的地表水质预测方法[D]. 哈尔滨: 哈尔滨工程大学, 2017.
[11]	赵文喜, 周滨, 刘红磊, 等.基于BP神经网络的海河干流叶绿素浓度短时预测研究[J]. 水利水电技术,2017,48(11):134-140.
[12]	张青, 王学雷, 张婷, 等.基于BP神经网络的洪湖水质指标预测研究[J]. 湿地科学,2016,14(2):212-218. ZHANG Q, WANG X L, ZHANG T, et al. Prediction of water quality index of Honghu Lake based on back propagation neural network model[J]. Wetland Science,2016,14(2):212-218.
[13]	赵林, 戴天骄, 陈亮, 等.基于BP神经网络的桃林口水库水质预测[J]. 安徽农业科学,2019,47(24):73-77. doi: 10.3969/j.issn.0517-6611.2019.24.024 ZHAO L, DAI T J, CHEN L, et al. Prediction of water quality in Taolinkou Reservoir based on BP neural network model[J]. Journal of Anhui Agricultural Sciences,2019,47(24):73-77. doi: 10.3969/j.issn.0517-6611.2019.24.024
[14]	刘双印. 基于计算智能的水产养殖水质预测预警方法研究[D]. 北京: 中国农业大学, 2014.
[15]	袁红春, 赵彦涛, 刘金生.基于PCA-NARX神经网络的氨氮预测[J]. 大连海洋大学学报,2018,33(6):808-813. YUAN H C, ZHAO Y T, LIU J S. Ammonia nitrogen level forecasting based on PCA-NARX neural network[J]. Journal of Dalian Ocean University,2018,33(6):808-813.
[16]	袁红春, 黄俊豪, 赵彦涛.基于PCA-NARX神经网络的溶解氧预测[J]. 山东农业大学学报(自然科学版),2019,50(5):902-907.
[17]	MULIA I E, TAY H, ROOPSEKHAR K, et al. Hybrid ANN-GA model for predicting turbidity and chlorophyll-a concentrations[J]. Journal of Hydro-Environment Research,2013,7(4):279-299. doi: 10.1016/j.jher.2013.04.003
[18]	王佳楠. 基于BP神经网络的高锰酸盐指数预测研究 : 以渭南潼关吊桥段为例[D]. 西安: 长安大学, 2017.
[19]	肖金球, 周翔, 潘杨, 等.GA-BP优化TS模糊神经网络水质监测与评价系统预测模型的应用: 以太湖为例[J]. 西南大学学报(自然科学版),2019,41(12):110-119.
[20]	LACHENBRUCH P A, COHEN J. Statistical power analysis for the behavioral sciences[J]. Journal of the American Statistical Association,1989,84(408):1096.
[21]	查木哈, 卢志宏, 翟继武, 等.双隐含层BP神经网络模型在老哈河水质预测中的应用[J]. 水资源与水工程学报,2018,29(2):56-61. ⊕ doi: 10.11705/j.issn.1672-643X.2018.02.10