蚁群算法及其在路径规划中的应用PPT学习课件

蚁群算法及其在路径规划中的应用,Q.Song2011.4.6,1,SwarmIntelligence,Dumbparts,properlyconnectedintoaswarm,yieldsmartresults.KevinKelly,2,Algorithmsinspiredfrominsects,AntColonyOptimization(ACO)ParticleSwarmOptimization(PSO)FishSwarm(FS),3,Twobridgesexperiment,fromnesttofoodconstrainedtomoveintwoasymmetricpaths,4,Principles,蚁群算法的基本原理源于昆虫学家们的观察和发现,生物界中的蚂蚁在搜索食物源时,能够在其走过的路径上释放一种蚂蚁特有的分泌物(pheromone)信息激素,使得一定范围内的其它蚂蚁能够觉察并影响其行为。当某些路径上走过的蚂蚁越来越多时,留下的这种信息激素也越多,以致后来蚂蚁选择该路径的概率也越高,从而增加了该路径的吸引强度,蚂蚁群体就是靠着这种内部的生物协同机制形成了一条它们自己并未意识到的最短路线。,5,Behaviorsofant,eachantmovesatrandom,pheromoneisdepositedonpath,antsdetectleadantspath,inclinedtofollow,morepheromoneonpathincreasesprobabilityofpathbeingfollowed,6,作为与遗传算法同属一类的通用型随机优化方法，蚁群算法不需要任何先验知识，最初只是随机地选择搜索路径，随着对解空间的“了解”，搜索变得有规律，并逐渐逼近直至最终达到全局最优解。蚁群算法对搜索空间的“了解”机制主要包括三个方面：(1)蚂蚁的记忆。一只蚂蚁搜索过的路径在下次搜索时就不会再被选择，由此在蚁群算法中建立tabu(禁忌)列表来进行模拟；(2)蚂蚁利用信息素进行相互通信。蚂蚁在所选择的路径上会释放一种叫做信息素的物质，当同伴进行路径选择时，会根据路径上的信息素进行选择，这样信息素就成为蚂蚁之间进行通讯的媒介。,7,(3)蚂蚁的集群活动。通过一只蚂蚁的运动很难到达食物源，但整个蚁群进行搜索就完全不同。当某些路径上通过的蚂蚁越来越多时，在路径上留下的信息素数量也越来越多，导致信息素强度增大，蚂蚁选择该路径的概率随之增加，从而进一步增加该路径的信息素强度，而某些路径上通过的蚂蚁较少时，路径上的信息素就会随时间的推移而蒸发。因此，模拟这种现象即可利用群体智能建立路径选择机制，使蚁群算法的搜索向最优解推进。,8,TheAlgorithm,UsedtosolveTSPTransitionfromcityitojdependson:-Tabulist:listofcitieshavingvisited-Visibility=1/dij;representslocalinformation,heuristicdesirabilitytovisitcityjwhenincityi.-Pheromonetrailforeachedge,representsthelearneddesirabilitytovisitcityjwhenincityi.,s,i,t,j,j,j,9,TheAlgorithm,TransitionRuleProbabilityofantkgoingfromcityitoj:Alphaandbetaareadjustableparametersthatcontroltherelativeimportanceoftrailversusvisibility;allowedk=Ntabuk,10,TheAlgorithm,Pheromoneupdate:,:acoefficientsuchthatrepresentstheevaporationoftrailbetweentimetandt+n,:thequantityperunitoflengthoftrailsubstancelaidonedge(i,j)byantkbetweentimetandt+n,11,TheAlgorithm,Threetypesof*ant-cyclesystem:Q:aconstantLk:thetourlengthofthekthant,12,TheAlgorithm,Threetypesof*ant-quantitysystem:,13,TheAlgorithm,Threetypesof*ant-densitysystem:,14,TheAlgorithm,Differencebetweenthreemodels:*ant-cyclesystemusesglobalinformation*ant-quantitysystemandant-densitysystemuseslocalinformation*ant-cyclesystemusedtosolveTSP,15,TSPExample,16,A,E,D,C,B,dAB=100;dBC=60;dDE=150,TSPExample,17,A,E,D,C,B,TSPExample,Iteration1,18,A,E,D,C,B,TSPExample,19,TSPExample,A,E,D,C,B,Iteration2,20,TSPExample,A,E,D,C,B,Iteration3,21,A,E,D,C,B,TSPExample,Iteration4,22,A,E,D,C,B,TSPExample,Iteration5,23,L1=300,L2=450,L3=260,L4=280,L5=420,TSPExample,24,EndofFirstRun,Allantsdie,Newantsareborn,SaveBestTour(Sequenceandlength),TSPExample,25,t=0;NC=0;ij(t)=cforij=0Placethemantsonthennodes,Updatetabuk(s),ComputethelengthLkofeveryantUpdatetheshortesttourfound=,Foreveryedge(i,j)ComputeFork:=1tomdo,Initialize,Choosethecityjtomoveto.Useprobability,Tabulistmanagement,Movek-thanttotownj.Inserttownjintabuk(s),Sett=t+n;NC=NC+1;ij=0,NCNCmax¬stagn.,Yes,End,No,Yes,26,VRPTWExample,NcustomersaretobevisitedbyKvehiclesGiven*Depots(number,location)*Vehicles(capacity,costs,timetoleave,timeonroad.)*Customers(demands,timewindows,priority,.)*RouteInformation(maximumroutetimeordistance,costontheroute),27,VRPTWExample,ObjectiveFunctionstoMinimize*Totaltraveldistance*Totaltraveltime*NumberofvehiclesSubjectto:*Vehicles(#,Capacity,timeonroad,triplength)*Depots(Numbers)*Customers(Demands,timewindows),28,VRPTWExample,RelationwithTSP?!,29,VRPTWExample,SimpleAlgorithm-Placeantsondepots(Depots#=Vehicle#).-Probabilisticchoice(1/distance,di,Q)amountofpheromone-Ifallunvisitedcustomerleadtoaunfeasiblesolution:Selectdepotasyournextcustomer.-Improvebylocalsearch.-Onlybestantsupdatepheromonetrial.,30,Advantages,PositiveFeedbackaccountsforrapiddiscoveryofgoodsolutions.Distributedcomputationavoidsprematureconvergence.Thegreedyheuristichelpsfindacceptablesolutionintheearlysolutionintheearlystagesofthesearchprocess.Thecollectiveinteractionofapopulationofagents.,31,