Page 17 - 2024年第55卷第3期
P. 17
[23] LIS,XINX,LEIZ.DynamicpathplanningofamobilerobotwithimprovedQ - learningalgorithm[C]??2015
IEEEInternationalConferenceonInformationandAutomation.IEEE ,2015.
[24] 时梦楠,崔博,王佳俊,等.复杂施工 条 件 下 无 人 碾 压 机 群 协 同 全 覆 盖 路 径 规 划 研 究 [J].水 利 学 报,
2020,51(12):1544 - 1557.
[25] SHIM G,WANGJJ,LIQH,etal.AcceleratedEarth - RockfillDamCompactionbyCollaborativeOperationof
UnmannedRollerFleet[J].JournalofConstructionEngineeringandManagement,2022,148(7).
[26] 林威伟,钟登华,胡炜,等.基于随机森林算法的土石坝压实质量动 态 评 价 研 究 [J].水 利 学 报,2018,
49(8):945 - 955.
[27] SUTTONR,BARTOA.ReinforcementLearning:AnIntroduction[M].Cambridge,MA:MITPress,1998.
[28] 李超,王瑞星,黄建忠,等.稀疏奖励下基于强化学习的无人集群自 主 决 策 与 智 能 协 同 [J].兵 工 学 报.
2023 ,44(6):1537 - 1546.
[29] MAHADEVANS,CONNELLJ.Automaticprogrammingofbehavior - basedrobotsusingreinforcementlearning
[J].ArtificialIntelligence,1992,55(2?3):311 - 365.
[30] WIEWIORAE,COTTRELLG,ELKANC.PrincipledMethodsforAdvisingReinforcementLearningAgents[M].
AAAIPress ,2003.
Dynamicpathplanningofrockfilldam warehousesurfacerollingoperationbasedonCA - RL
CUIBo,ZHONGHang,WANGJiajun,TANTianwen,LINWeiwei,TONGDawei
(StateKeyLaboratoryofHydraulicEngineeringIntelligentConstructionandOperation,TianjinUniversity,Tianjin 300350,China)
Abstract:Rollingoperationisessentialtotheconstructionofrockfilldam.Scientificandreasonableplanningof
therollingpathofwarehousesurfacecanimprovetherollingefficiencyandmaintaintherollingquality.Currentre
searchonrollerpathplanningduringstorehousesurfacescompactionlacksin - depthconsiderationofdynamic
factorssuchaschangesinthenumberofrollingmachinesandreal - timeanalysisofcompactionquality.Regarding
thissituation ,thispaperproposesadynamicpathplanningmethodforrollingmachinegroupsbasedonreinforce
mentlearning - instructedcellularautomatamodelinstructed.First ,acellularautomata - basedrollingsurfaceinfor
mationmodelisestablished ,andamethodforevaluatingtheoverallcompactionqualityofstripsisproposedto
storeandupdatethecompactionqualityandotherwarehousesurfaceinformation.Then ,apathplanningmodel
basedonreinforcementlearningisestablished ,thestatesetandactionsetareconstructed,therewardfunctionis
designedandtheutilizationstrategyisexploredtosolvethepathassignmentproblem asthenumberofrollingma
chineschanges.Coupledwiththeabovetwomodels ,thedynamicpathplanningofrollergroupsinrockfilldam
constructionisrealized.Theengineeringapplicationshowsthattheproposedmethodcandynamicallyconsider
changingfactorssuchasnumberofrollersandperceptionofcompactionquality.Theplannedpathreducesinlength
by22.3% onaveragecomparedwithon - siteconstructionwhilemaintaininghighcompactionquality.Theproposed
methodcansignificantlyimprovetherollingefficiency.
Keywords:rockfilldamconstruction;warehousesurfacerollingoperation;dynamicpathplanning;cellularautom
ata ;reinforcementlearning
(责任编辑:李福田)
— 2 6 5 —