文档名：自动化立体仓库退库货位优化问题及其求解算法
摘要：针对自动化立体仓库出库作业过程中剩余货物退库问题,以堆垛机作业总能耗最小化为目标,以退库货位分配为决策变量,建立了自动化立体仓库退库货位优化模型,提出了基于深度强化学习的自动化立体仓库退库货位优化框架.在该框架内,以立体仓库实时存储信息和出库作业信息构建多维状态,以退库货位选择构建动作,建立自动化立体仓库退库货位优化的马尔科夫决策过程模型;将立体仓库多维状态特征输入双层决斗网络,采用决斗双重深度Q网络(duelingdoubledeepQ-network,D3QN)算法训练网络模型并预测退库动作目标价值,以确定智能体的最优行为策略.实验结果表明D3QN算法在求解大规模退库货位优化问题上具有较好的稳定性.
Abstract：Theefficientretrievalhandlingofstorageproductsinautomatedstorageandretrievalsystem(AS/RS)isthekeyforjust-in-timeproductioninindustrialproductionsystems.Inaproduction-orientedAS/RS,theamountofaspecificproducttoberetrievedisoftenlessthanapalletload.Oncompletionofeachretrievalrequest,theremainingproductsonthepalletshouldberelocatedtoaspecificracklocation.Incurrentpractice,afterretrievingcertainitems,anon-emptypalletistypicallyreturnedtotheoriginallydesignatedlocation.Itisofsignificantimportancetodynamicallyre-assigningthestoragelocationsofnon-emptypalletsaftereachretrievalinordertoimproveretrievaloperationalefficiencyaswellasenergyconsumption.Inthisstudy,aretrieval-orientedstoragerelocationproblemisinvestigatedtominimizeenergyconsumption.Basedonthecharacteristicsoftheaccelerationanddecelerationmotionofthestackercraneandthecrane'sdualoperationmode,anenergyconsumptionmodelofAS/RSduringretrievalprocessesisfirstestablished.Thentherelocationofnon-emptypalletaftereachretrievalrequestistakenasanoperationaldecisionandtheconstraintsincludingrackstabilityandmaximumstoragecapacityofeachracklocationareconsidered,andanoptimizationmodelofretrieval-orientedstoragerelocationisbuiltwiththeobjectiveofminimizingthetotalenergyconsumptionduringtheretrievalprocesses.InconsiderationoftheNP-hardcharacteristicoftheoptimizationmodel,thispaperproposesanoptimizationframeworkbasedondeepreinforcementlearningalgorithm.Withinthisframework,themulti-dimensionalstateisdesignedbasedonthereal-timestorageinformationaswellastheretrievalinformationandtheactionisconstructedbytheretrieval-orientedstoragerelocation.AMarkovdecisionprocessmodelforretrieval-orientedstoragerelocationisthenestablishedandtheduelingdoubledeepQ-networks(D3QN)algorithmisemployedtoobtaintheoptimalstoragerelocationofnon-emptypalletaftereachretrievalrequest.InD3QNalgorithm,themulti-dimensionalstateofAS/RSisputintotheduelingnetworktopredictmainQvalueaswellastargetQvalueandthenetworktrainingmechanisminthedoubledeepQnetworkalgorithmisintroducedtotrainbothmainnetworkandtargetnetwork.Whenthenetworkmodelconvergestotheoptimalvaluefunction,theintelligentagentfindsanoptimalracklocationforthenon-emptypalletaftereachretrievalrequestbasedonthereal-timestateofAS/RS,soastominimizeenergyconsumptiondemandedbythestackercrane.Finally,theexperimentalcasesofdifferentscalesaredesignedunderthevariabilityofrackshapesandproductretrievalfrequency.Asetofsensitiveanalysisexperimentsareconductedtodeterminethehyper-parametersofD3QN,includingtraininglearningrate,thesizeoftrainingbatch,thenumberofneuronsinthehiddenlayerandnetworkupdatefrequency.Toevaluatetheeffectivenessoftheproposedalgorithm,extensivecomparisonsbetweenD3QN,DDQNandDQNaremadeunderdifferentcases.OurexperimentalresultsshowD3QNalgorithmobtainsbettersolutionperformanceindealingwiththelarge-scaleretrieval-relatedrelocationcases.Moreover,acomparativeanalysisoftheproposedrelocationmethodandthemethodofreturningtotheoriginalstoragelocationisconductedunderdifferentcases.Ourexperimentalresultsshowtheproposedrelocationmethodsignificantlyreducestotalenergyconsumptionduringretrievalprocesses.Ourstudyconsidersthesequenceofretrievalrequestsasgivenratherthandecisionvariables.Ifretrievalsequencingisincorporatedwithretrieval-orientedstoragerelocation,furtherimprovementonretrievalefficiencyandenergyconsumptionmaybeachieved.Thus,furtherinvestigationsareneededontheintegratedoptimizationofretrieval-orientedstoragerelocationandretrievalsequencingforAS/RS.
作者：何在祥 李丽 张云峰 郗琳Author：HEZaixiang LILi ZHANGYunfeng XILin
作者单位：西南大学工程技术学院,重庆400715
刊名：重庆理工大学学报 PKU
Journal：JournalofChongqingInstituteofTechnology
年，卷(期)：2024, 38(5)
分类号：TP18
关键词：自动化立体仓库 退库货位优化 深度强化学习 D3QN
Keywords：automatedstorageandretrievalsystem retrieval-orientedstoragerelocation deepreinforcementlearning duelingdoubledeepQ-networks
机标分类号：TP301.6O241.6F273
在线出版日期：2024年5月24日
基金项目：重庆市杰出青年科学基金项目，中央高校基本科研业务费专项资金项目自动化立体仓库退库货位优化问题及其求解算法[
期刊论文] 重庆理工大学学报--2024, 38(5)何在祥 李丽 张云峰 郗琳针对自动化立体仓库出库作业过程中剩余货物退库问题,以堆垛机作业总能耗最小化为目标,以退库货位分配为决策变量,建立了自动化立体仓库退库货位优化模型,提出了基于深度强化学习的自动化立体仓库退库货位优化框架.在该...参考文献和引证文献
参考文献
引证文献
本文读者也读过
相似文献
相关博文
自动化立体仓库退库货位优化问题及其求解算法 Retrieval-oriented storage relocation problem of an automated storage and retrieval system and its solving algorithm
自动化立体仓库退库货位优化问题及其求解算法.pdf
- 文件大小:
- 5.28 MB
- 下载次数:
- 60
-
高速下载
|
|
