Page 89 - 2024年第55卷第11期
P. 89
[16] SHANGY,LUS,GONGJ,etal.Improvedgeneticalgorithm foreconomicloaddispatchinhydropowerplants
andcomprehensiveperformancecomparisonwithdynamicprogrammingmethod [J].JournalofHydrology,2017,
554:306 - 316.
[17] 周建中,李英海,肖舸,等.基于混合粒子群算法的梯级水电站多目标优化调度 [J].水 利 学 报,2010,
41(10):1212 - 1219.
[18] 赵雪花,黄强,吴建华.蚁群算法 在 水 电 站 厂 内 经 济 运 行 中 的 应 用 [J].水 力 发 电 学 报,2009,28(2):
139 - 142,138.
[19] 陈立华,梅亚东,董雅洁,等.改进遗传算法及其在水库群优化调度中的应用[J].水利学报,2008,39
(5):550 - 556.
[20] SUTTONRS,BARTOAG.ReinforcementLearning:AnIntroduction[M].MITpress,2018.
[21] CASTELLETTIA, PIANOSIF, RESTELLIM.A multiobjective reinforcementlearning approach to water
resourcessystemsoperation:Paretofrontierapproximationinasinglerun[J].WaterResourcesResearch,2013,
49(6):3476 - 3486.
[22] 李文武,刘江鹏,蒋志强,等.基于 HSARSA( λ )算法的水库长期随机优化调度研究[J].水电能源科学,
2020,38(12):53 - 57.
[23] JIANGW Y,LIUYQ,FANGGH,etal.Researchonshort - termoptimalschedulingofhydro - wind - solarmulti -
energypowersystem basedondeepreinforcementlearning [J].JournalofCleanerProduction,2023, 385.doi:
10.1016?j.jclepro.2022.135704.
[24] ZENGY,WENX,TANQ,etal.Real - timeloaddispatchinhydropowerplantbasedonD3QN - PER[J].Jour
nalofHydrology ,2023,625:130019.
[25] MNIH V,KAVUKCUOGLUK,SILVERD,etal.Human - levelcontrolthroughdeepreinforcementlearning[J].
Nature,2015,518(7540):529 - 533.
ResearchonoptimalloadallocationinhydropowerstationbasedonDQNalgorithm
1
1,2
1,2
1
TANQiaofeng ,SONGJiawei,WENXin ,ZENGYuxuan,WANGHao 3
(1.HohaiUniversity,CollegeofWaterConservancy&HydropowerEngineering,Nanjing 210098,China;
2.HohaiUniversity,CooperativeInnovationCenterforWaterSafety&Hydro - Science,Nanjing 210098,China;
3.ChinaInstituteofWaterResourcesandHydropowerResearch,Beijing 100038,China)
Abstract:Loadallocationofhydropowerstationisthebasisofautomaticpowergenerationcontrolsystem,andthe
conventionalmethodisdifficulttoadapttothecomplexanddiversifiedmulti - objectiveregulationandcontrolre
quirements.Thispaperconstructsaloaddistributionmodelthattakesintoaccountthedemandofwaterresources
andelectricpower ,anddevelopsanefficientsolutionalgorithmbasedontheDQN(DeepQ - Netwoek)algorithm.
TakingZhentoubaHydropowerStationasanexample ,thewaterconsumptionofpowergenerationinthedryperiod
isreducedby0.14%,whichimprovestheoperationeconomyofthepowerstation;andthevarianceofwaterlevel
2
variationinthefloodseasonisreducedby0.876m ,whicheffectivelysuppressesthewaterlevelfluctuationofthe
powerstation.Atthesametime ,theaccuracyofloaddistributioncalculationisimprovedfrom3 - 5MW to1MW,
andthecalculationefficiencyisimprovedby88.2 - 106.4times,whichcanprovidetechnicalsupportforthesched
ulingandoperationofcomplexhydropowersystemsinthenewperiod.
Keywords:loadallocation;reinforcementlearning;economicoperation;water - electricityregulation;efficient
decisionmaking
(责任编辑:王 婧)
3
— 1 5 5 —