A MultiAgent Based Coordination Approach for Improving the Energy Efficiency In Large SN.docx

A MultiAgent Based Coordination Approach for Improving the Energy Efficiency In Large SN.docx

《A MultiAgent Based Coordination Approach for Improving the Energy Efficiency In Large SN.docx》由会员分享，可在线阅读，更多相关《A MultiAgent Based Coordination Approach for Improving the Energy Efficiency In Large SN.docx（33页珍藏版）》请在冰豆网上搜索。

AMultiAgentBasedCoordinationApproachforImprovingtheEnergyEfficiencyInLargeSN

AMulti-AgentBasedCoordinationApproachforImprovingtheEnergyEfficiencyInLargeSensorNetwork

YuLasheng1,LiJie2

1SchoolofInformationScienceandEngineering,CentralSouthUniversity,China

2DepartmentofComputerScience,graduateSchoolofSystemsandInformationEngineering,UniversityofTsukuba,Japan

（ley462yu@ ,lijie@cs.tsukuba.ac.jp）

1.Introduction

Increasingsocietaldemandforautomationhasledtoconsiderableeffortstocontrollarge-scalecomplexsystems,especiallyintheareaofautonomousintelligentcontrolmethods.Sensornetworks,havingthepotentialtorevolutionizethecapturing,processing,andcommunicationofcriticaldataforuse,havebeendevelopedfortheserequirements.Asthenamesuggests,Asensornetworkconsistsofanumberofsensornodes,eachsensornodeisequippedwithoneormoresensorsthatproducemeasurementsoftheenvironmentandsomecomponentsincludingasmallcomputer,anenergysourcesuchasabattery,andaradiotransmitter–receiver（transceiver）.Itrepresentsthenextstepinwirelesscommunication’sminiaturization,andthesensornodes’powerandsizemakeitfeasibleforawiderangeapplicationsfromscientificapplicationssuchasdisastersurveillancetoEngineeringapplicationssuchasmilitary,industrialproductlinemonitoring,agriculturalandwildlifeobservation,healthcare,smarthomes,etc.Theseapplicationsemployalargenumberofsensornodesforcontinuoussensinganddatagathering.Eachsensorperiodicallyproducesasmallamountofdataandreportstocertainremotesinknode（orbasestation）inanautonomousandunattendedmanner.Butregardlessoftheapplication,itisdesirabletominimizethecostofthesystem,tomaximizethecoverageandenergyefficiencyofthewholesystem.

Energyefficiencyofthewholesystemisoneoftheprimarychallengestothesuccessfulapplicationofwirelesssensornetworks（WSNs）sincethesensornodesarepoweredwithlimitedbatteriesandtheycannotbeeasilyrechargedoncedeployed.Evenso,thenodesmusttobeautonomous,andsurviveonphysicalenvironmentalconditionsbeingasenergy-efficientaspossible.Uptonow,manyenergyefficientroutingalgorithmsorprotocolshavebeenproposedwithtechniqueslikeclustering,dataaggregation,multi-pathandlocationtracking,etc.,ascanbeseenfromrelatedwork.However,manyofthemaimtominimizeparametersliketotalenergyconsumptionordelayduringtheroutingprocess,whichmightcausesomehotspotnodesaswellasapartitionednetworkduetotheoveruseofcertainnodes.

Toadapttheserequirements,aWSNneedstoadoptsmartapproaches,comprisingflexibility,autonomyinmanagingtheobservedenvironmentevents,withoutcompromisinghigherenergyconsumption.ManysmartproposalsforoptimizationofenergyconsumptiononthenodesarealreadypossibletobefoundindifferentpartsfromWSNstackprotocols,suchas:

puttingthenodestosleepduringidleperiods;settingshortduty-cycle;andperformingin-networkprocessingtoavoiddataredundancytransmission.Nevertheless,thereisnoautonomyinsuchapproaches,andtheapplicationlayerstillusuallyactasasimpleslave,only,storing　andforwardingitsenvironmentreadings,doingonlywhatwassaidtobedone,withoutconsideraboutwhatitwouldliketo.

However,inmanyapplicationscenarios,thesystemwideperformanceofthenetworkisdependentnotonlyonthedutycyclesoftheindividualsensors,butalsoontheinteractionsbetweenthesesensors.Theserequirementsgiverisetovariousoptimizationproblemssuchasautonomyoftheseapproacheswhichcanbeenforcedbasedonstaticandmobileagentssystems.

Inthispaper,theproblemrepresentationwasfirstpresentedandthenanintelligentarchitecture,anenergy-efficientmulti-agentbasedarchitecturewhichmainlyfocusesonimprovingtheenergyefficiencyofthewholesensornetworkhasbeenproposed.Inourproposals,eachsensorwasrepresentedbyanagent（infactasensoragentandapossibledeliveragent）withastateandautilityfunction,whichcanoptimizetheutilityofthesysteminadecentralizedwaytoachievegoodenergyefficiencyandnetworkperformancebasedonthecoordinationamongagents,.Inparticular,weuseanalgorithmbasedonmax-sumalgorithmwhichallowsagentstooptimizeadecomposableglobalfunctionthroughdistributedlocalcomputationandcommunication.Thealgorithmproposedinthispaperisabletocomputesolutionsveryclosetotheoptimal,exhibitsalowmessageoverhead,isextremelyrobusttomessagefailures,andiscapableofbeingdeployedonlow-powersensingdevices

Therestofthepaperisorganizedasfollows:

Insection2,therelatedworkisbrieflyreviewed.Insection3,themulti-agentarchitecturewiththreeagents,classificationagent,erroragentandfilteragentwereintroduced.Insection4,thecoordinationmechanismsamongagentswerediscussed.Insection5,theimplementationsforthecoordinationsbasedonminimumspanningtree’salgorithmhavebeendiscussedindetail.Insection6,thesimulationresultsbasedonthemulti-agentarchitecturehasbeengivenandanalyzed.Insection7,asummaryofthemainideasanddirectionsforfutureworkhasbeenconcluded.

2Relatedwork

Theproblemofimprovingtheenergyefficiencyandbalancingtheenergyconsumptionetchasbeenconsideredpreviouslyintheliterature.Oneoftheearliestpapers　onthetopicistheworkofRamanathanandRosales-Hainproposedacentralizedalgorithmforfindingtheminmax　transmissionpowerlevelwhichaddressesthe　probleminthesettingofmaximizingthelifetimeofasingle-sessionbroadcast.Theirapproaches,however,　weresuboptimalanddidnotguaranteeagoodresultinallcases.　KangandPoovendrandiscussedseveralproblemsrelatedtodynamiclifetime　maximizationsuchastheissueofnon-uniformenergylevelsrelyingondistributedmethodsfor　constructingminimumspanningtrees.Theydidnotemphasize　the　minimumtotalenergymetricforthepurposeofmaximizingnetworklifetime.Narayanaswamyetal.proposedaprotocolforpowercontrolinwirelessadhoc　networkswithdiscretepowerlevels,whichalsoestablishedaspanning　subgraphofthetransmissiongraphwithminmaxcost.However,thesolutionmainlyreliedonaproactiveroutingprotocolandrequiressignificantcontroloverhead　thatisunsuitableforsensornetworks.RodopluandMengpresentedadistributedalgorithmbasedontheconceptofrelayregions:

eachnodeisawareofitsowngeographic　locationandthelocationofitsneighbors,eachnode　canlocallydeterminewhichneighboritshouldforwardthemessagetoinorderto　minimizethetotalenergyconsumption.Thisapproachwasoptimalbut　requiresextensiveassumptions,suchastheavailabilityoflocationinformationandconstructinga　specificpath-lossmodel.　Wattenhoferetal.proposedasimilardistributedalgorithmwhichreliesonageometriccone-basedforwardingscheme,requiringthat　nodescouldmeasureexactlythedirectionofincomingradiotransmissions（e.gangleof　arrival）.Guo,Yang,andLeungproposedadistributedalgorithm　DistributedMin-MaxTree（DMMT）toconstructamulticasttreeswithminimum　maximumtransmissioncost,followingPrim’salgorithmforconstructingminimum　spanningtrees,thepurposeistomaximizethesensornetworklifetime.

Totackletheproblemofdynamicallychangingenvironments,coordinationmechanismappliedtoalargedistributedsensornetworkhasextensivelybeenstudied.Manyapproachesarebasedondecentralizedapproacheswherethedecisionsforthepositionandfunctionofasensororagroupofsensorsistakenlocallyrelyingontheinformationcollectedfromthenearneighborhood.Anotabledecentralizedapproachistorepresenteachsensorasanautonomousagentwhichcaninteractonlywithnearbysensor-agents.Thecoordinationbetweenthesensorsthenisbasedonmulti-agentcoordinationmethodologies.AtypicalapproachwasproposedbyGrocholskyetal.Theyhavedistinguishedcooperativesolutionsbasedonexplicitnegotiationandcoordinatedsolutionsbasedonlyonobservedinformationexchange.ThisapproachusedDistributedDataFusionalgorithmsandoptimizedassignmentswithrespecttoinformation-theoreticobjectivefunctions.Itemploysadistributedoptimizationalgorithm,however,itrequiresasmoothanddifferentiableobjectivefunction.Inpractice,itisoftenrequiredtointroduceheuristics（suchasasynchronousupdates,noisetermsandrandomizedupdateorders）topreventoscillationsandhelpescapeunstableequilibriumorweaklocaloptimumsolutions.Horlingetal.provideasummaryofMASarchitecturesandalgorithmsappliedtosensorresourceallocation.Patricioet.al.haveproposedamulti-agentframeworkforsurveillanceapplicationswherethesurveillanceprocessisdistributedoverseveralsensoragents,accordingtotheirindividualabilitytocontributetheirlocalinformationtotheglobaltargetsolution.Farinelliat.al.,andStranderset.al.haveproposedthedecentralizedcoordinationofmultipleagentswiththehelpofthegraphtheoreticmethodofMaxSumalgorithm.Bymaximizingthesocialwelfareofthegroupofagentstheycanoptimizethesense/sleepcyclesofasetofwirelesssensorsandpathplanningformonitoringandpredictingthestateofspatialphenomena.Theseapproachesarehighlyreliantontheshapeandnatureoftheobjectivefunctionthatcannotbeguaranteedbecauseitishighlydependentonthesensormodelused.

TheMaxSumalgorithmisakindoftheGeneralized　DistributiveLawframework.ItprovidesgoodqualityapproximatesolutionstoDistributedConstraintOptimizationsPr