《9、李林-湖泊藻类水华的HEA模拟预警(29页).pdf》由会员分享,可在线阅读,更多相关《9、李林-湖泊藻类水华的HEA模拟预警(29页).pdf(29页珍藏版)》请在三个皮匠报告上搜索。
1、湖泊藻类水华的遗传算法模型预测研究第五届中国水环境模型与智能决策研讨会.玉溪.2018.10.25李李 林、林、闪锟、闪锟、宋立荣宋立荣Hongqing Cao,Friedrich Recknagel主主要要内内容容13一、研究背景一、研究背景二二、遗传算法简介遗传算法简介三三、HEA模型用于藻类水华预测的案例模型用于藻类水华预测的案例3藻类水华预测的数据驱动模型方法 人工神经网络(Artificial Neural Network,ANN)(Recknagel,F.,French,M.,Harkonen,P and K.Yabunaka 1997;Maier HR,Dandy GC,Burc
2、h MD,1998;Liu Y,Yao X,1999;Lee JHW,Fernando TMKG,Wong KTM,2004)4杂交演化算法(Hybrid Evolutionary Algorithm,HEA)基于时间序列水质参数的规则设定的水华藻类基于时间序列水质参数的规则设定的水华藻类HEA模拟原理图模拟原理图*Cao,H.,Recknagel,F.,Kim,B.and N.Takamura(2006)研究目的 能否用常规易检测的理化参数来预测湖库的藻能否用常规易检测的理化参数来预测湖库的藻类水华?类水华?HEAHEA模型预测藻类水华的效果如何?模型预测藻类水华的效果如何?与与ANNANN
3、方法相比,方法相比,HEAHEA模型有何优势,能否筛选模型有何优势,能否筛选关键的参数及其阈值?关键的参数及其阈值?滇池位置:云南昆明容量:15亿方平均水深:4 m面积:300 km2流域面积:2920 km2Haigeng Bay1.滇池藻类水华的滇池藻类水华的HEA模拟预测模拟预测Variation of Total Phosphorus includes three stages:TP rose slowly from 1960 to 1990;TP was accelerated from 1991 to 2000;TP was fallen down since 2001.Varia
4、tion of Total Nitrogen includes two stages:Total nitrogen steady rise from 1960 to 2007;TN started to fall since 2008.滇池水质变化滇池水质变化TP TN 滇池外海叶绿素的变化滇池外海叶绿素的变化滇池藻类组成变化(滇池藻类组成变化(2008-2010)phylumgenusspecies滇池藻类滇池藻类HEA模拟的各参数变化范围模拟的各参数变化范围VariablesNameMeanMinMaxInput VariablesWater Temperature oC(WT)18.88
5、.029.5Secci Depth m(SD)26.3570Diss Oxygen mg/L(DO)8.83.216.0pH9.17.210.3Chemical Oxygen Demand mg/L(COD)28.81050Ammonia NH4-N mg/L(NH4-N)0.4802Filt Nitrite NO2-N mg/L(NO2-N)0.3800.5Filt Nitrate NO3-N mg/L(NO3-N)0.5602.1Total Nitrogen mg/L(TN)3.20.813Filt Reac Phosphorus mg/L(PO4-P)0.03200.16Total Ph
6、osphorus mg/L(TP)0.260.031.1Wind Direction(WD)6.7012Wind Speed m/s(WS)2.009.2Output VariablesChlorophyll-a ug/L(Chla)12052300Cyanophyceae mg/L870.03956Chlorophyceae mg/L8.30120Bacillariophyceae mg/L 2.3068Aphanizomenon mg/L10.10563Anabaena mg/L3.90186Microcystis mg/L29.70.02538Historical DataReal-ti
7、me in situWater QualityMeasurementsby Hydrolab DataSonde 5X Water Temperature CDissolved Oxygen mg/lpHTurbidity NTUAmmonium NH4 mg/lTotal Chlorophyll g/lData Acquisitionby HydrolabProcessMonitorData Merger and ValidationOnline DataLake Data Warehouse LDWEarly Warning for Operational Raw Water Contro
8、l if Cyanobacteria Bloom is ImminentTime SeriesChlorophyll-a g/lEvolutionaryModelling Real-Time ForecastingDays/Weeks AheadModelData Preprocessing ModuleForecasting ModuleConductivitymS/cmPhycocyanoin g/lOnline DataA.circinalis cells/mlC.raciborskii cells/mlCylindrospermopsin g/lHEA模拟预测的流程图输入水质理化参数:
9、COD,DO,NH4,NO2,NO3,pH,SD,SRP,TN,TP,Water temperature 气象数据:风速,风向时间:2008.10-2010.9Output variables:Chla,Biomass of Algae(phylum,gerena,species)Hybrid Evolutionary Algorithms:输出建模00500600700800Chla7d-predicted_ChlaIF(WT14.648)Chla=(TN*(TP*107.892)Chla=(WT-(TP*WT)*(TP*(55.408-WD)-WD)THENELSEH
10、EAHEA预测滇池叶绿素的最佳模型预测滇池叶绿素的最佳模型Best Model005006007008002008/10/12008/11/12008/12/12009/1/12009/2/12009/3/12009/4/12009/5/12009/6/12009/7/12009/8/12009/9/12009/10/12009/11/12009/12/12010/1/12010/2/12010/3/12010/4/12010/5/12010/6/12010/7/12010/8/12010/9/1Chla7d-predicted_ChlaIF(WT14.648)Chla=
11、(TN*(TP*107.892)Chla=(WT-(TP*WT)*(TP*(55.408-WD)-WD)THENELSE叶绿素叶绿素a a的的HEAHEA预测模型的灵敏度分析预测模型的灵敏度分析0500050100TP:0.24-0.45WT:13.46-22.71WD:4.20-9.2900500600700050100COD:30.99-42.53TN:5.75-7.40TP:0.59-0.81Sensitivity Analysis15Input Variables Selection frequency0554045CO
12、D NH4 NO2 NO3 pHSDSRPTNTPWT WD WS05008/10/12008/12/12009/2/12009/4/12009/6/12009/8/12009/10/12009/12/12010/2/12010/4/12010/6/12010/8/1Water Temperature(Celsius)Water Temperature(Celsius)0500300350400450Chlorophyll-a(ug/L)Chlorophyll-a(ug/L)y=14.059x-84.848R=0.5520025
13、03003504004500102030Chlorophyll-a(ug/L)Chlorophyll-a(ug/L)线性(Chlorophyll-a(ug/L)从预测模型看水温对叶绿素的影响从预测模型看水温对叶绿素的影响WT005006007008002008/10/12008/11/12008/12/12009/1/12009/2/12009/3/12009/4/12009/5/12009/6/12009/7/12009/8/12009/9/12009/10/12009/11/12009/12/12010/1/12010/2/12010/3/12010/4/12010/
14、5/12010/6/12010/7/12010/8/12010/9/1Chla7d-predicted_Chla3d-predicted_Chla14d-predicted_ChlaItemselected best ruleset modeltotal errortotal R23d-Predicted ChlaIF(WT14.648)38.16620.90THEN Chla=(TN*(TP*107.892)ELSE Chla=(WT-(TP*WT)*(TP*(55.408-WD)-WD)7d-Predicted ChlaIF(WT=13.720)39.28480.89THEN Chla=(
15、WT*(15.728-WD)ELSE Chla=(TP*(81.224-ln(|NO3|)*WD)14d-Predicted ChlaIF(WT14.648)38.16620.90THEN Chla=(TN*(TP*107.892)ELSE Chla=(WT-(TP*WT)*(TP*(55.408-WD)-WD)3d-Predicted ChlaIF(WT=13.720)39.28480.89THEN Chla=(WT*(15.728-WD)ELSE Chla=(TP*(81.224-ln(|NO3|)*WD)14d-Predicted ChlaIF(WT=15.415)43.18300.87
16、THEN Chla=(exp(pH)-70.340)/64.110)+exp(ln(|WS|)ELSE Chla=(exp(pH)-313.306)/96.217)+(exp(TN)-(TP*(-550.157)/3.8)HEA 预测模型选择的水温数值与野外监测中Chla明显增加时的水温值非常接近,说明HEA模型具备选择关键参数及其阈值的潜力。HEAHEA预测模型选择参数的意义预测模型选择参数的意义WT=14.8主要藻门类生物量提前7天的预测结果05003003502008/10/12008/12/12009/2/12009/4/12009/6/12009/8/12009
17、/10/12009/12/12010/2/12010/4/12010/6/12010/8/1Cyanopredicted_Cyano007080902008/10/12008/12/12009/2/12009/4/12009/6/12009/8/12009/10/12009/12/12010/2/12010/4/12010/6/12010/8/1Chloropredicted_Chloro058/10/12008/12/12009/2/12009/4/12009/6/12009/8/12009/10/12009/12/12010/2/12010/4/
18、12010/6/12010/8/1Bacillapredicted_BacillaItemThe best ruleset model for 7d-ahead-predictiontotal errortotal R2蓝藻IF(SD146.831)OR(SD43.869)THENChloro=(WS*(pH/46.933)*SD)ELSEChloro=(ln(|(TN*17.252)|)*71.637)/(TP*(WS*SD)+COD)4.34130.91硅藻IF(exp(TN)=45.808)AND(WT=15.490)AND(SD=30.427)AND(DO=105.824)THENAn
19、abeana=(TP*(DO/WS)*(TP*147.343)-(WD+85.129)+ln(|NH4|)ELSE Anabeana=exp(exp(TP-NO3)/(NH4+TP)2.85220.98AphanizomenonIF(SRP+SD)=(-33.546)OR(SD=20.730)AND(SD=37.746)OR(WT20.154)THENT-Microcystis=(123.704-(COD/TP)ELSEMicrocystis=(TN-SD)-(TP/(TN-65.070)/exp(pH)15.11090.9002040608008/10/12009/10
20、/1M.wesenbergiipredicted_M.wesenbergii0554045502008/10/12008/12/12009/2/12009/4/12009/6/12009/8/12009/10/12009/12/12010/2/12010/4/12010/6/12010/8/1M.novacekiipredicted_M.novacekii002008/10/12008/12/12009/2/12009/4/12009/6/12009/8/12009/10/12009/12/12010/2/12010/4/12010/6/12010/
21、8/1M.viridispredicted_M.viridis024681012142008/10/12008/12/12009/2/12009/4/12009/6/12009/8/12009/10/12009/12/12010/2/12010/4/12010/6/12010/8/1M.aeruginosapredicted_M.aeruginosa常见微囊藻种类生物量的常见微囊藻种类生物量的7天预测天预测ItemThe best ruleset model for 7d-ahead-predictiontotal errortotal R2M.wesenbergiiIF(WT=17.750)
22、AND(SD146.464)OR(DO=12.290)AND(SD=19.662)OR(pH9.772)THEN M.viridis=(ln(|(COD/WD)|)+(ln(|(-73.683)/WD)|)+SD)ELSE M.viridis=(TP*(TN*4.493)+(COD/SD)+ln(|exp(NH4)|)5.23520.81M.novacekiiIF(WT22.693)THEN M.aeruginosa=(TP*(TP*(NO2*14.534)*WD)ELSEM.aeruginosa=(WS*57.755)*SRP*SRP)*(WS*ln(|SD|)*SRP)0.79310.90
23、常见微囊藻种类生物量的提前常见微囊藻种类生物量的提前7天预测的天预测的HEA模型模型模型及预警阈值设定3 Days Forecast ModelBest 1Best 2Best 3ThresholdAnabaena(mg/L)YYY10Aphanizomenon(mg/L)YYY15Microcystis(mg/L)YYY20Chlorophyll-a(ug/L)YYY307 Days Forecast ModelBest 1Best 2Best 3ThresholdAnabaena(mg/L)YYY10Aphanizomenon(mg/L)YXX15Microcystis(mg/L)YYY2
24、0Chlorophyll-a(ug/L)XXX3014 Days Forecast ModelBest 1Best 2Best 3ThresholdAnabaena(mg/L)YYY10Aphanizomenon(mg/L)YYY15Microcystis(mg/L)YYY20Chlorophyll-a(ug/L)YYY30小小 结结 杂交演化算法杂交演化算法HEA得到的滇池得到的滇池Chla、硅藻、绿藻、蓝藻、主要蓝藻属、硅藻、绿藻、蓝藻、主要蓝藻属(Anabeana、Aphanizomenon、Microcystis)和微囊藻的主要组成种类)和微囊藻的主要组成种类(Microcystis
25、novacekii、M.viridis、M.wesenbergii、M.aeruginosa)的预测模)的预测模型结果均较好(型结果均较好(r2 0.80)。)。提前提前3天、天、7天、天、14天的预测模型结果显示,预测时间越短,预测模型精度越天的预测模型结果显示,预测时间越短,预测模型精度越高。高。HEA 预测模型选择的水温数值与野外监测中预测模型选择的水温数值与野外监测中Chla明显增加时的水温值非常接明显增加时的水温值非常接近,说明近,说明HEA模型具备选择关键参数及其阈值的潜力。可望模型具备选择关键参数及其阈值的潜力。可望为管理部门控制为管理部门控制蓝藻水华危害提供决策信息支持。蓝藻水
26、华危害提供决策信息支持。Wivenhoe Reservoir,QueenslandLocation:upper Brisbane RiverMax.Volume:1165 GLMax.Depth:79mCatchment Area:7020 km2大坝大坝2.澳大利亚澳大利亚WivenhoeWivenhoe水库水库Cylindropermopsis 的HEAHEA模拟模拟Cylindropermopsis 的HEAHEA预测模型预测模型Cylindropermopsis 的HEAHEA预测模型灵敏度分析预测模型灵敏度分析结 语1.将HEA用于三个湖库的不同水华藻类生物量的预测都获得了较好的效果
27、。从HEA模型的r2来看,滇池水华蓝藻的预测模型最高(r20.80),其次是以色列Kinneret 湖甲藻(PeridiniumPeridinium)水华预测模型(0.64 r2 0.76),澳Wivenhoe水库Cylindropermopsis 预测模型略低(0.59 r2 0.64)。这可能与不同湖泊富营养化程度、水华藻种类及生物量不同有关。2.HEA模型有助于筛选影响藻类水华的关键参数及其阈值,如水温等,对水华预警与控制管理具有积极作用。致致 谢谢 滇池水专项课题(2008-2017)EES,University of Adelaide NSFC-ISF中国以色列国际合作项目(2015-2018)谢谢!敬请指正!李李 林林 中国科学院水生生物研究所中国科学院水生生物研究所藻类生物学与应用研究中心藻类生物学与应用研究中心