Page 76 - 2022年第53卷第8期
P. 76
[35] 冯平,商颂,李新.基于分位数回归的滦河流域降水及径流变化特性[J].水力发电学报,2016,35(2):
28 - 36.
[36] ZHANGZ,YEL,QINH,etal.Windspeedpredictionmethodusingsharedweightlongshort - termmemorynet
workandgaussianprocessregression[J].AppliedEnergy,2019,247:270 - 284.
[37] HEF,ZHOUJ,FENGZ,etal.Ahybridshort - termloadforecastingmodelbasedonvariationalmodedecompo
sitionandlongshort - term memorynetworksconsideringrelevantfactorswithBayesianoptimizationalgorithm[J].
AppliedEnergy,2019,237:103 - 116.
[38] 金俊?,武鹏,董祥祥,等.基于 RF - GRU风速预测的风电 MPPT控制[J].传感器与微系统,2021,40
(5):38 - 41.
[39] CHENT,GUESTRINC.XGBoost:Ascalabletreeboostingsystem[C].AssocCompMachinery,2016.785 - 794.
Aprobabilisticforecastingframeworkoftimeseriesvariablesfor
wind - solar - hydropowerhybridsystems
1,2
3
1,2
1,2
ZHANGZhendong ,LUOBin ,QINHui,TANGHaihua ,
1,2
1,2
ZHOUChao ,FENGKuaile
(1.ChangjiangSurvey,Planning,DesignandResearchCo.,Ltd,Wuhan 430010,China;
2.Internet +SmartWaterConservancyKeyLaboratory,ChangjiangWaterResourcesCommittee,Wuhan 430010,China;
3.SchoolofCivilandHydraulicEngineering,HuazhongUniversityofScienceandTechnology,Wuhan 430074,China)
Abstract:Thereal - timedispatchofwind - solar - hydropower(WSH)hybridsystemisaffectedbytheuncertainty
oftime - seriesvariablessuchaswindspeed,solarradiationintensity,runoff,andpowerload.Howtoaccurately
forecastthesevariablesandquantifytheuncertaintyisthekeyproblemfacedbytheWSHhybridsystem.Inorderto
solvetheproblem ,aprobabilisticforecastingframeworkfortimeseriesvariablesbasedonadeeplearningmodelis
proposedbythisstudy.First ,thefeatureinputisminedfromthetimeseriesdataandthecorrelationcoefficientis
usedtoselectthegeneratedfeatures.Then ,basedondeeplearningmodelandGaussianprocessregression,the
timeseriesvariableprobabilisticforecastingmodelisconstructed,andthefeaturecombinationoptimizationandhy
perparameteroptimizationarerealizedthroughthe0 - 1planningideaandtheBayesianoptimizationalgorithm re
spectively.Theforecastingmodeliscomprehensivelyevaluatedfrom threeaspects: deterministicforecasting,
probabilisticforecastingandreliability.Finally,takingtheWSH complementationpilotdemonstrationbaseinthe
YalongRiverBasinastheresearchobject ,theframeworkproposedinthisstudyiscomparedwiththecurrentseven
state - of - the - arttime - seriesvariableforecastingmodelsonthefourdatasetsofrunoff ,windspeed,photovoltaics
andpowerload,respectively,toverifytheaccuracyandprobabilisticcomprehensiveperformance.
Keywords:wind - solar - hydropowerhybridsystem;probabilisticforecasting;deeplearning;featurecombination
optimization ;hyperparameteroptimization
(责任编辑:于福亮)
— 9 6 3 —