基于NARX神经网络的千岛湖藻类短期预测模型构建

李未; 朱梦圆; 王裕成; 朱广伟

doi:10.13198/j.issn.1001-6929.2022.02.17

基于NARX神经网络的千岛湖藻类短期预测模型构建

doi: 10.13198/j.issn.1001-6929.2022.02.17

1.
中国科学院南京地理与湖泊研究所，湖泊与环境国家重点实验室，江苏南京　210008
2.
杭州市生态环境局淳安分局，浙江杭州　311700

基金项目: 国家自然科学基金项目(No.41977339)；中国科学院野外站联盟项目(No.KFJ-SW-YW036)；江苏省水利科技项目(No.2020004)

详细信息

作者简介:
李未（1980-），女，山东临清人，副研究员，博士，主要从事湖库蓝藻水华短期预测预警研究，liwei@niglas.ac.cn

中图分类号: X524
计量
- 文章访问数: 178
- HTML全文浏览量: 97
- PDF下载量: 51
- 被引次数: 0
出版历程
- 收稿日期: 2021-11-30
- 修回日期: 2022-02-12

Short-Term Forecasting Model for Algae Based on NARX Neural Network in Qiandaohu Reservoir

1.
State Key Laboratory of Lake Science and Environment, Nanjing Institute of Geography and Limnology, Chinese Academy of Sciences, Nanjing 210008, China
2.
Hangzhou Bureau of Ecology and Environment Chun'an Branch, Hangzhou 311700, China

Funds: National Natural Science Foundation of China (No.41977339)；Field Station Alliance Project of Chinese Academy of Sciences, China (No.KFJ-SW-YW036)；Water Resource Science and Technology Project in Jiangsu Province, China (No.2020004)

摘要

摘要: 局部水域的藻类异常增殖现象逐渐成为千岛湖面临的水环境保护难题. 构建以数据驱动的水华预测模型，实现对重点水域叶绿素a (Chla)浓度短期动态变化的预测，是快速应对潜在水华风险的有效手段之一. 鉴于NARX神经网络在预测非平稳时间序列动态特征方面的优势，以千岛湖国控监测断面小金山2016—2019年Chla的高频时间序列作为研究对象，对Chla剖面数据进行沿深平均、缺失值插补后，分别以连续3 d和连续7 d的Chla浓度作为输入，构建了基于NARX神经网络的藻类预测模型，用于预测未来0.5~7 d Chla浓度的变化，探讨了相关参数设置、训练及评价方法，并针对不同的预见期分析了模型性能. 结果表明：① 模型预测性能稳定，预测值与实测值相关系数保持在0.8~0.9之间，均方误差在15~30之间. ②随着预见期的变化，模型性能不同. 其中，在未来0.5~4 d的预测中，使用连续3 d的 Chla浓度作为输入的预测效果较好；在未来4.5~7 d的预测中，使用连续7 d的Chla浓度作为输入的预测效果较好. 研究显示，该模型可以较为准确地预测未来0.5~7 d的Chla浓度，可为构建以数据驱动的千岛湖水华监测预警系统提供科学依据.
- 藻类预测 /
- 高频序列 /
- 神经网络 /
- 千岛湖
Abstract: In recent years, Qiandaohu Reservoir has suffered from abnormal growth of algae. Short-term forecasting models of algal blooms driven by data can be built to predict the temporal variation of chlorophyll-a (Chla) concentration in some concern water. It is considered effective to deal with potential bloom risks. NARX (nonlinear autoregressive with external input) neural network has a competitive advantage in predicting the dynamic characteristics of non-stationary time series. In this paper, a high-frequency monitoring buoy dataset of Xiaojinshan Station, a state-controlled section in Qiandaohu Reservoir from 2016 to 2019 was handled with vertical integration and missing value interpolation. An algal bloom forecasting model based on NARX neural network was then built to predict the temporal variations of Chla over the next 0.5-7 days. To drive the model, Chla concentrations from the first 3 and 7 days were used as initial inputs, respectively. Parameters settings, trainings and evaluations were discussed, and model performance was analyzed according to different foreseeable periods. The results showed that: (1) The performance of the model was steady. The correlation coefficients between prediction and observation reached 0.8-0.9, and the corresponding mean square error values were 15-30. (2) In the 0.5-4 days of forecasting, the model using the first 3-day Chla concentration as the initial input had higher accuracy, while the model using the first 7-day chlorophyll concentration as initial input showed better precision in the forecast for days 4.5-7. It is therefore recommended to us models with different inputs for different forecast periods. The model contributed to provide a methodology for a bloom early warning system driven by data.
- algae forecasting /
- high-frequency time series /
- neural network /
- Qiandaohu Reservoir

HTML全文

图 1 千岛湖库体及国控监测断面分布

Figure 1. Qiandaohu Reservoir and location of state-controlled sections

下载: 全尺寸图片幻灯片

图 2 小金山国控断面高频监测Chla浓度的剖面变化情况

Figure 2. High frequency observed Chla profile at Xiaojinshan Staion

下载: 全尺寸图片幻灯片

图 3 NARX神经网络拓扑结构

注：X(t)为外部输入向量；Y(t)为期望目标向量；d为延时阶数；L和b分别为神经网络各层的权值矩阵及偏置，其中，上标1代表隐含层，上标2代表输出层；f为各层激活函数，其中，隐含层的激活函数f ¹取双曲正切函数，输出层的激活函数f ²取线性函数.

Figure 3. Topology architecture of NARX neural network

下载: 全尺寸图片幻灯片

图 4 两类模型对未来0.5 d Chla浓度的预测值与观测值对比

Figure 4. Performance of two NARX models forecasting the dynamics of Chla in the future 0.5 day

下载: 全尺寸图片幻灯片

图 5 两类模型对未来3 d Chla浓度的预测值与观测值对比

Figure 5. Performance of two NARX models forecasting the dynamics of Chla in the future 3 day

下载: 全尺寸图片幻灯片

图 6 两类模型对未来6 d Chla浓度的预测值与观测值对比

Figure 6. Performance of two NARX models forecasting the dynamics of Chla in the future 6 day

下载: 全尺寸图片幻灯片

表 1 两类模型预测性能对比

Table 1. Prediction performance of two NARX models

预测时间/d	第1类模型		第2类模型
预测时间/d	均方误差	相关系数	均方误差	相关系数
0.5	14.873 9	0.906 0	17.970 0	0.889 1
1	17.802 1	0.895 6	21.203 8	0.872 0
1.5	21.668 5	0.860 6	21.252 3	0.895 1
2	30.179 9	0.821 7	19.472 3	0.875 8
2.5	20.040 9	0.883 5	26.882 3	0.861 5
3	18.671 5	0.893 3	25.329 5	0.850 4
3.5	19.054 2	0.879 9	26.371 2	0.842 0
4	27.698 6	0.830 2	30.232 8	0.848 3
4.5	25.332 1	0.856 2	23.903 1	0.846 9
5	22.195 0	0.856 4	24.383 0	0.870 7
5.5	28.066 7	0.850 9	38.656 5	0.777 4
6	27.782 3	0.824 2	11.819 7	0.929 4
6.5	22.477 7	0.882 9	23.565 6	0.870 7
7	30.385 5	0.813 9	23.425 4	0.873 8

下载: 导出CSV

参考文献(40)

[1]	盛海燕,吴志旭,刘明亮,等.新安江水库近10年水质演变趋势及与水文气象因子的相关分析[J].环境科学学报,2015,35(1):118-127. SHENG H Y,WU Z X,LIU M L,et al.Water quality trends in recent 10 years and correlation with hydro-meteorological factors in Xin'anjiang Reservoir[J].Acta Scientiae Circumstantiae,2015,35(1):118-127.
[2]	笪文怡,朱广伟,吴志旭,等.2002—2017年千岛湖浮游植物群落结构变化及其影响要素[J].湖泊科学,2019,31(5):1320-1333. doi: 10.18307/2019.0522 DA W Y,ZHU G W,WU Z X,et al.Long-term variation of phytoplankton community and driving factors in Qiandaohu Reservoir,southeast China[J].Journal of Lake Sciences,2019,31(5):1320-1333. doi: 10.18307/2019.0522
[3]	吴志旭,刘明亮,兰佳,等.新安江水库(千岛湖)湖泊区夏季热分层期间垂向理化及浮游植物特征[J].湖泊科学,2021,24(3):460-465. WU Z X,LIU M L,LAN J,et al.Vertical distribution of phytoplankton and physico-chemical characteristics in the lacustrine zone of Xin'anjiang Reservoir (Lake Qiandao) in subtropic China during summer stratification[J].Journal of Lake Sciences,2021,24(3):460-465.
[4]	韩晓霞,朱广伟,吴志旭,等.新安江水库(千岛湖)水质时空变化特征及保护策略[J].湖泊科学,2013,25(6):836-845. doi: 10.18307/2013.0607 HAN X X,ZHU G W,WU Z X,et al.Spatial-temporal variations of water quality parameters in Xin'anjiang Reservoir (Lake Qiandan) and the water protection strategy[J].Journal of Lake Sciences,2013,25(6):836-845. doi: 10.18307/2013.0607
[5]	LI Y,ZHANG Y,SHI K,et al.Spatiotemporal dynamics of chlorophyll-a in a large reservoir as derived from Landsat 8 OLI data:understanding its driving and restrictive factors[J].Environmental Science and Pollution Research,2018,25(2):1359-1374. doi: 10.1007/s11356-017-0536-7
[6]	笪文怡,黎云祥,朱广伟,等.水文气象过程对千岛湖氮磷变化的影响[J].水生态学杂志,2019,40(5):9-19. DA W Y,LI Y X,ZHU G W,et al.Influence of hydrometeorological processes on nutrient dynamics in Qiandao Lake[J].Journal of Hydroecoloty,2019,40(5):9-19.
[7]	LIU M,ZHANG Y,SHI K,et al.Spatial variations of subsurface chlorophyll maxima during thermal stratification in a large,deep subtropical reservoir[J].Journal of Geophysical Research:Biogeosciences,2020,125(2):e2019JG005480.
[8]	韩博平.中国水库生态学研究的回顾与展望[J].湖泊科学,2010,22(2):151-160. HAN B P.Reservoir ecology and limnology in China:a retrospective comment[J].Journal of Lake Sciences,2010,22(2):151-160.
[9]	史鹏程,朱广伟,杨文斌,等.新安江水库悬浮颗粒物时空分布、沉降通量及其营养盐效应[J].环境科学,2020,41(5):2137-2148. SHI P C,ZHU G W,YANG W B,et al.Spatial-temporal distribution of suspended solids and its sedimentation flux and nutrients effect in Xin'anjiang Reservoir,China[J].Environmental Science,2020,41(5):2137-2148.
[10]	LIU M,ZHANG Y,SHI K,et al.Thermal stratification dynamics in a large and deep subtropical reservoir revealed by high-frequency buoy data[J].Science of the Total Environment,2019,651:614-624. doi: 10.1016/j.scitotenv.2018.09.215
[11]	笪文怡,朱广伟,黎云祥,等.新安江水库河口区水质及藻类群落结构高频变化[J].环境科学,2020,41(2):713-727. DA W Y,ZHU G W,LI Y X,et al.High-frequency dynamics of water quality and phytoplankton community in inflowing river mouth of Xin'anjiang Reservoir,China[J].Environmental Science,2020,41(2):713-727.
[12]	LIU M,ZHANG Y,SHI K,et al.Effects of rainfall on thermal stratification and dissolved oxygen in a deep drinking water reservoir[J].Hydrological Processes,2020,34:3387-3399.
[13]	ZHANG M,ZHANG Y,DENG J,et al.High-resolution temporal detection of cyanobacterial blooms in a deep and oligotrophic lake by high-frequency buoy data[J].Environmental Research,2022,203:111848. doi: 10.1016/j.envres.2021.111848
[14]	ROUSSO B Z,BERTONE E,STEWART R,et al.A systematic literature review of forecasting and predictive models for cyanobacteria blooms in freshwater lakes[J].Water Research,2020,182:115959. doi: 10.1016/j.watres.2020.115959
[15]	刘天,姚梦雷,黄继贵,等.BP 神经网络在传染病时间序列预测中的应用及其MATLAB实现[J].预防医学情报杂志,2019,35(8):812-821. LIU T,YAO M L,HUANG J G,et al.Application of back propagation neural network in prediction of infectious disease time series and its MATLAB implementation[J].Journal of Preventive Medicine Information,2019,35(8):812-821.
[16]	张怡文,胡静宜,王冉.基于神经网络的PM_2.5预测模型[J].江苏师范大学学报(自然科学版),2015,33(1):63-65. ZHANG Y W,HU J Y,WANG R.PM_2.5 forecasting model based on neural network[J].Journal of Jiangsu Normal University(Natural Science Edition),2015,33(1):63-65.
[17]	赵文怡,夏丽莎,高广阔,等.基于加权KNN-BP神经网络的PM_2.5浓度预测模型研究[J].环境工程技术学报,2019,9(1):14-18. doi: 10.3969/j.issn.1674-991X.2019.01.003 ZHAO W Y,XIA L S,GAO G K,et al.PM_2.5 prediction model based on weighted KNN-BP neural network[J].Journal of Environmental Engineering Technology,2019,9(1):14-18. doi: 10.3969/j.issn.1674-991X.2019.01.003
[18]	SEO Y,KIM S,KISI O,et al.Daily water level forecasting using wavelet decomposition and artificial intelligence techniques[J].Journal of Hydrology,2015,520:224-243. doi: 10.1016/j.jhydrol.2014.11.050
[19]	屈雅静,魏海英,马瑾.基于BP神经网络的北京城区公园土壤PAHs含量预测[J].环境科学研究,2020,33(12):2864-2871. QU Y J,WEI H Y,MA J.Prediction of polycyclic aromatic hydrocarbons (PAHs) content in soil of urban parks in Beijing based on BP neural network[J].Research of Environmental Sciences,2020,33(12):2864-2871.
[20]	任加国,龚克,马福俊,等.基于BP神经网络的污染场地土壤重金属和PAHs含量预测[J].环境科学研究,2021,34(9):2237-2247. REN J G,GONG K,MA F J,et al.Prediction of heavy metal and PAHs content in polluted soil based on BP neural network[J].Research of Environmental Sciences,2021,34(9):2237-2247.
[21]	曾纳,任小丽,何洪林,等.基于神经网络的三江源区草地地上生物量估算[J].环境科学研究,2017,30(1):59-66. ZENG N,REN X L,HE H L,et al.Aboveground biomass of grasslands in the three-river headwaters region based on neural network[J].Research of Environmental Sciences,2017,30(1):59-66.
[22]	刘恒,严力蛟.BP神经网络在千岛湖水体富营养化变化预测中的应用[J].科技通报,2008,24(3):411-416. doi: 10.3969/j.issn.1001-7119.2008.03.024 LIU H,YAN L J.Back-propagation network model for predicting the change of eutrophication of Qiandao Lake[J].Bulletin of Science and Technology,2008,24(3):411-416. doi: 10.3969/j.issn.1001-7119.2008.03.024
[23]	XIAO X,HE J,HUANG H,et al.A novel single-parameter approach for forecasting algal blooms[J].Water Research,2017,108:222-231. doi: 10.1016/j.watres.2016.10.076
[24]	代兴兰.遗传算法与最小二乘支持向量机在年径流预测中的应用[J].水资源与水工程学报,2014,25(6):231-235. doi: 10.11705/j.issn.1672-643X.2014.06.049 DAI X L.Application of genetic algorithm and least squares support vector machine in prediction of annual runoff[J].Journal of Water Resources and Water Engineering,2014,25(6):231-235. doi: 10.11705/j.issn.1672-643X.2014.06.049
[25]	崔东文,金波.基于改进的回归支持向量机模型及其在年径流预测中的应用[J].水力发电学报,2015,34(2):7-14. CUI D W,JIN B.Improved support vector machine regression model and its application to annual runoff forecasting[J].Journal of Hydroelectric Engineering,2015,34(2):7-14.
[26]	SIEGELMANN H T,HORNE B G,GILES C L.Computational capabilities of recurrent NARX neural networks[J].IEEE Transactions on Systems,Man & Cybernetics:Part B,1997,27(2):208.
[27]	ZOUNEMAT-KERMANI M,STEPHAN D,HINKELMANN R.Multivariate NARX neural network in perdiction gaseous emissions within the influent chamber of wastewater treatment plants[J].Atmospheric Pollution Research,2019,10(6):1812-1822. doi: 10.1016/j.apr.2019.07.013
[28]	ROGHANCHI P,KOCSIS K C.Quantifying the thermal damping effect in underground vertical shafts using the nonlinear autoregressive with external input (NARX) algorithm [J].International Journal of Mining Science and Technology,2019,29(2):255-262.
[29]	WONG C X,WORDEN K.Generalised NARX shunting neural network modeling of friction[J].Mechanical Systems and Signal Processing,2007,21(1):553-572. doi: 10.1016/j.ymssp.2005.08.029
[30]	SHOAIB M,SHAMSELDIN A Y,MELVILLE B W,et al.A comparison between wavelet based static and dynamic neural network approaches for runoff prediction[J].Journal of Hydrology,2016,535:211-225. doi: 10.1016/j.jhydrol.2016.01.076
[31]	赵琦琳,邱飞,杨健.NARX神经网络模型在昆明市环境空气质量预测中的应用[J].中国环境监测,2019,35(3):42-48. ZHAO Q L,QIU F,YANG J.Application of NARX neural network model in environmental air quality prediction in Kunming[J].Environmental Monitoring in China,2019,35(3):42-48.
[32]	张丹宁,张猛,张博.基于NARX神经网络的PM_2.5/10浓度值预测模型:以咸阳市两寺渡监测站为例[J].地球环境学报,2020,11(2):161-168. ZHANG D N,ZHANG M,ZHANG B.The model to predict PM_2.5/10 concentrations based on NARX neural network:taking Liangsidu monitoring station in Xianyang as an example[J].Journal of Earth Environment,2020,11(2):161-168.
[33]	刘墨阳,李巧玲,李致家,等.基于小波分析的NARX神经网络在水位预测中的应用[J].南水北调与水利科技,2019,17(5):56-63. LIU M Y,LI Q L,LI Z J,et al.The application of NARX neural network model based on wavelet analysis for water level prediction[J].South-to-North Water Transfers and Water Science & Technology,2019,17(5):56-63.
[34]	范哲南,刘小生.NARX神经网络在大坝变形预测中的应用[J].人民黄河,2021,43(2):18-21. FAN Z N,LIU X S.Application of NARX neural network in dam deformation prediction[J].Yellow River,2021,43(2):18-21.
[35]	方苏阳,蒋创,魏涛,等.基于时间序列的动态神经网络沉降预测[J].测绘与空间地理信息,2018,41(11):24-27. doi: 10.3969/j.issn.1672-5867.2018.11.007 FANG S Y,JIANG C,WEI T,et al.Time series-dynamic neural network on settlement prediction[J].Geomatics & Spatial Information Technology,2018,41(11):24-27. doi: 10.3969/j.issn.1672-5867.2018.11.007
[36]	孙国祥,闫婷婷,汪小旵,等.基于小波变换和动态神经网络的温室黄瓜蒸腾速率预测[J].南京农业大学学报,2014,37(5):143-152. doi: 10.7685/j.issn.1000-230.2014.05.023 SUN G X,YAN T T,WANG X C,et al.A method of cucumber transpiration rate forecast based on wavelet transform and dynamic neural network[J].Journal of Nanjing Agricultural University,2014,37(5):143-152. doi: 10.7685/j.issn.1000-230.2014.05.023
[37]	陈婷.NARX动态神经网络的择时策略研究[D].上海:上海师范大学,2019.
[38]	GUZMAN S M,PAZ J O,TAGERT M L M.The use of NARX neural networks to forecast daily groundwater levels[J].Water Resources Management,2017,31(5):1591-1603. doi: 10.1007/s11269-017-1598-5
[39]	YUE S,PILON P,CAVADIAS G.Power of the Mann-Kendall and Spaman's test for detecting monotonic trends in hydrological series[J].Journal of Hydrology,2002,259:254-271. doi: 10.1016/S0022-1694(01)00594-7
[40]	ZHAO Y,KOCKELMAN K M.The propagation of uncertainty through travel demand models:an exploratory analysis[J].Annals of Regional Science,2002,36:145e163.