多agent系統(tǒng)中并行聯(lián)盟機(jī)制研究.pdf

1、合肥工業(yè)大學(xué)博士學(xué)位論文多agent系統(tǒng)中并行聯(lián)盟機(jī)制研究姓名：蘇兆品申請(qǐng)學(xué)位級(jí)別：博士專業(yè)：計(jì)算機(jī)應(yīng)用技術(shù)指導(dǎo)教師：蔣建國(guó)20081001ABSTRACTInopen，distributed，andheterogeneousmultiagentsystems(MAS)，allagentmaybemotivatedtointeractandcooperatewithotheragentswhenitencounterstasksthat

2、aredifficulttoaccomplishwithitsnarrowfunctionalcapabilitiesandresourcesMoreover’itisoftenbeneficialtoassignagroupofagentstoataskandthatgroup&agentsisusuallycalleda“coalition”Formingcoalitionsbecomesnecessarywhenasingleag

3、entcannotperformataskResearch011coalitioninMASisamajorresearchchallengeinthefieldofcontroltheoryHowevermostexistingresearchesareaddressedin“singletaskcoalition’and“serialgenerationofmultitaskcoalitions”，limitingthescopeo

4、ftheirapplicationsinreal—worldscenariosAgainstthisbackground，wedevelopanovel“ParallelCoalition’inenvironmentswheremanytasksmustbeexecutedsimultaneouslyInthisdissertation，weaddresstheproblemofgeneration，formationandevalua

5、tionofparallelcoalitioninmulti—taskenvironmentswhereanagentmayjoininseveraldifferentcoalitionsOuraimistoimprovetheefficiencyofsolvingtasksandutilizingresourcesandprovideatheoreticalandmethodologicalguidanceformanycomplic

6、atedcontrolproblemsThemainresearchcontentsandinnovativecontributionsofthisdissertationareaSfollows：(1)Weproposeamulti—taskcoalitionparallelgenerationalgorithmbasedonimprovedimlnunealgorithmAnabilitypartitionstrategybased

7、ondimensionandanovel‘‘ChildAgent”areintroducedAnimprovedimmunealgorithmwithvaccineadaptiveobtainingandthreedimensionalbinaryencodingisappliedtosolvemultitaskcoalitionparallelgenerationproblemSpeciallyouralgorithmrealizes

8、theconditionthatanagentCantakepartinseveraldifferentcoalitionssimultaneouslyMoreovertheexperimentalresultshowsthatouralgorithmisefficientandrobust(2)Weproposeamultitaskcoalitionparallelformationstrategybasedonreinforceme

9、ntlearningFirstlywededucefromtheorythattheprocessofmulti—taskcoalitionformationisaMarkovdecisionprocessThenreinforcementlearningmechanismisadoptedtodescribeagents。behaviorstrategyandtheprocessofmulti—taskcoalitionparalle

10、lformationisdescribedInparallelmultitaskdomains，ourstrategyformsmultitaskcoalitionseffectivelyandCallconvergetoalloptimalsolutionwithprobabilityone(3)WeproposeanoveltwolayeredevaluationmethodonagentcoalitionInthisdissert

11、ationweevaluateacoalitionfromuniversalityandaggregateallexpertsconclusionbasedonDempsterShafer(D—S)evidentialtheorywhichcallclearlydescribeuncertaintyandincompletionofproblemsandexpressexperts’comprehensionsmartlyWeevalu

12、ateonagentcoalitionaccordingtoagentabilityharmoniousperformance，communicationcost，familiarityandsustainabledevelopmentSpeciallyourevaluationmethodcompletelyaccordswithhumant11inkingandjudgment。Moreovegourmethodisagile，va