基于群智能的复杂联盟机制研究

《基于群智能的复杂联盟机制研究》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、合肥工业大学博士学位论文基于群智能的复杂联盟机制研究姓名：张国富申请学位级别：博士专业：计算机应用技术指导教师：蒋建国20080501ABSTRACTThedistributedintelligentcontrolbasedonmulti—agentsystems(MaS)isspringingupvigorouslyforanapplicationtOcooperativetaskssupportedbycomputers．Therefore，researchoncoalitioninthefieldofMASisbecoming

2、moreimportant．Formingstableandbalancedcoalitionsisamajor托=seaI℃hchallengeinthefieldofcontroltheory．Existingresearchessufferfromallimportantdrawbackthatanagentcallnotbutjoininonlyacoalitionandacoalitioncannotbutbeengagedinonlyatask，producingabigwasteofresourcesandcapabi

3、lities，andlimitingthescopeoftheirapplicationsinreal-worldscenarios．Againstthisbackground,wedevelopanovel‘'ComplicatedCoalition’’andadoptswarmintelligencetoaddresstheproblemofcoalitionformationinmultitaskenvironmentswhereallagentmayjoininseveraldifferentcoalitionsandaco

4、alitionmayperformseveraldifferenttaskssimultaneously．Ouraimistoimprovetheefficiencyofsolvingtasksandutilizingresourccsandprovideatheoreticalandmethodologicalguidanceformanycomplicatedcontrolproblems．Themainresearchcontentsandinnovativecontributionsofthisdissertationare

5、asfollows：(1)Wedevelopaparticleswarmscooperativeoptimizationbasedalgorithmforcomplicatedcoalitionserialgeneration．Anovel‘'V'n．tualAgent'’enlightenedbygraphtheoryisusedtOtransfertheremaindercapabilitiesofafathercoalition．Tothelattertasks，thefathercoalitionwilldelegateit

6、spowertothevirtualagenttocompetefortasks．Specially,ouralgorithmrealizestheconditionthatanagentcarltakepartinseveraldifferentcoalitionsandacoalitionCanturnitshandtOseveraldifferenttasks，partlyreducingthewasteofresourcesandcapabilities．Moreover,theexperimentalresultshows

7、thatOuralgorithmismoreefficientformoretasks、析msmallcapabilities．(2)Wedevelopdiscreteparticleswarmoptimizationwithtwo-dimensionalbinaryencodingforsolvingcomplicatedcoalitionparallelgeneration．Specially,checkingonencodingvalidityandstrategiesforconflictresolutionarebroug

8、htintoeffecttosurmountresourceconflictandcoalitionlockwhenseveraldifferentcoalitionscompeteagainsteachotherforthesame