欢迎来到天天文库
浏览记录
ID:39401771
大小:386.00 KB
页数:69页
时间:2019-07-02
《基于Q学习的Agent智能决策的研究与实现》由会员上传分享,免费在线阅读,更多相关内容在行业资料-天天文库。
1、合肥工业大学硕士学位论文基于Q学习的Agent智能决策的研究与实现姓名:虞靖靓申请学位级别:硕士专业:计算机应用技术指导教师:王浩20050401pcKroeomyvDweldeotxrhddaotmsQa:liMeaAmSi,ngRmoebmoodCcuanpe,丘Q宅c1tei锄viel1y1i9mp,mCve血eimelligenceofna.gentdecisioninMAC,co叩erationTheResearchandImplementationofAgentIntelIigentDecisionBasedonQLearningABSTRACTResearchont11et
2、11eor)ra11d印plicationofmulti-agentsystem(MAS)hasbecomeahotspotofanmcialintelligence.111eRobotWorldCup(RoboCup)isatypicalMAswithcharacterssuchasdyn锄icenviro眦ent,tllecoexistenceofcooperationa11dcompetitionamongseveralagents,limitedcommunicationband嘶dtll,atldt11enoisyenViromem.Basedonmisgene蹦tcstpl甜
3、.oⅡn,varioustlleoriesaIldalgorithmsofMAScallbereseafchedandevaluated.Reinfbrcement1e蛐ingisanunsupervisedlearningtecllllology,bywhichtheagemc趾findoptimalbehaviorsequenceaIldperfbmon—line1eaming.soreinforcementleamingisrecogllizedasanideaItechnolog)rtoconstI!uctiIltelligentagent.In血isdissena_tion,m
4、eapproachforagenttomakeintelligentdecisionisstudiedbasedonQ1eamingmemod,includingtlledesignofdecision丘啦ework孤dtlleimplementationofindividualskillalldte锄c00peration.Considcringthccomplexi母ofagcⅡtdecisiontaskinRoboCup,layerleamingbasedondecision行锄eworkisdesigned.nle丘arneworkdiVides恤fulldccisiontask
5、intoseverallayersfromhigll—levelt0low・level.Machinele唧ing(ML)tecllIliquesareusedineachlayerandmehi曲er1ayerisimplementedbasedontllelower1ayer.Thelayercddecisionf}锄e、Ⅳorkcombincd、ⅣitllMLtechniquescallOvercomethelimitationofhand・codedimplemen洲on.Inordertoimprovemeintelligenceofi11divid砌s枷s,Qle啪ingme
6、modisadoptedtotrainb雒icskillsofballkicl【inga11ddribbling.SincctllestatespaceinRoboCupiscontinuous,CMACnetworkisusedtorealizctllestategcneralizationinQleaming.Fortlleleanlingproblemofagentte锄cooperation,mebasicQleamingalgorithInisextendedbyintroducingtlleconceptof1eadmgagent锄d∞mr培也ercwardassignmen
7、tpolicyamongseve豫IleamingagentS.TWotypicalteamcooperationproblermssuchasbalIpass锄d2vslcooperationaresolVedbymeextendedQle锄ingalgonhIn.AllmecxperimentsaremadeunderRoboCupsimulationplatfb珊.Theresultshave合肥工业大学本论文经答辩委员会全体
此文档下载收益归作者所有