sensors Article An Improved Simulated Annealing Technique for Enhanced Mobility in Smart Cities HayderAmer1,*,NaveedSalman1,MatthewHawes1,MoumenaChaqfeh2, LyudmilaMihaylova1andMartinMayfield3 1 DepartmentofAutomaticControlandSystemsEngineering,UniversityofSheffield,SheffieldS13JD,UK; n.salman@sheffield.ac.uk(N.S.);m.hawes@sheffield.ac.uk(M.H.);l.s.mihaylova@sheffield.ac.uk(L.M.) 2 CollegeofInformationTechnology,UAEUniversity,AlAin15551,UnitedArabEmirates; [email protected] 3 DepartmentofCivilandStructuralEngineering,UniversityofSheffield,SheffieldS13JD,UK; martin.mayfield@sheffield.ac.uk * Correspondence:hmamer1@sheffield.ac.uk;Tel.:+44-114-222-5675 AcademicEditors:AndreaZanellaandToktamMahmoodi Received:24April2016;Accepted:24June2016;Published:30June2016 Abstract: Vehiculartrafficcongestionisasignificantproblemthatarisesinmanycities. Thisisdueto theincreasingnumberofvehiclesthataredrivingoncityroadsoflimitedcapacity. Thevehicular congestion significantly impacts travel distance, travel time, fuel consumption and air pollution. Avoidanceoftrafficcongestionandprovidingdriverswithoptimalpathsarenottrivialtasks. The keycontributionofthisworkconsistsofthedevelopedapproachfordynamiccalculationofoptimal trafficroutes. Twoattributes(theaveragetravelspeedofthetrafficandtheroads’length)areutilized bytheproposedmethodtofindtheoptimalpaths. Theaveragetravelspeedvaluescanbeobtained fromthesensorsdeployedinsmartcitiesandcommunicatedtovehiclesviatheInternetofVehicles androadsidecommunicationunits. Theperformanceoftheproposedalgorithmiscomparedtothree otheralgorithms: thesimulatedannealingweightedsum, thesimulatedannealingtechniquefor orderpreferencebysimilaritytotheidealsolutionandtheDijkstraalgorithm. Theweightedsum andtechniquefororderpreferencebysimilaritytotheidealsolutionmethodsareusedtoformulate differentattributesinthesimulatedannealingcostfunction. AccordingtotheSheffieldscenario, simulationresultsshowthattheimprovedsimulatedannealingtechniquefororderpreferenceby similaritytotheidealsolutionmethodimprovesthetrafficperformanceinthepresenceofcongestion by an overall average of 19.22% in terms of travel time, fuel consumption and CO emissions as 2 comparedtootheralgorithms;also,similarperformancepatternswereachievedfortheBirmingham testscenario. Keywords: traffic congestion; Internet of Vehicles; Internet of Things; simulated annealing; multi-objectiveoptimisation;vehiclere-routing 1. Introduction Theexplosivegrowthoftheglobaleconomyhasledtoanexpansionofcities. Therehasbeen anincreaseinthepopulationmassand,therefore,anincreaseinthenumberofvehiclesdrivingon cityroadnetworksoflimitedcapacity. Thishaspromptedanextremeincreaseoftrafficcongestion, roadaccidentsandairpollution. Somestudieshaverevealedthat30%ofCO emissionsaredueto 2 inefficientvehiclemanagement[1]. Thishasresultedinsignificanteconomicandproductivitylosses, makingimprovementofmobilityakeychallengewithinsmartcities. Solvingsuchaproblemcanbe aidedbyinformationobtainedviasensorsdeployedaspartofsmartcityinitiatives,whichthenhave tobecommunicatedtovehicles/driverstoallowthemtomakeadecisionwithregardstoalternative Sensors2016,16,1013;doi:10.3390/s16071013 www.mdpi.com/journal/sensors Sensors2016,16,1013 2of23 routes. Oncesuchadecisionhadbeenmade,itwouldalsobefeasibletoeventuallyseeinformation regardingplannedroutestobecommunicatedbackfromthevehiclestothesmartcityinfrastructure. Suchinformationcan thenbeusedto predictthe numberofvehiclesat each intersectionwithina smartcity,whichinturncouldbeusedtoadaptthesequencesoftrafficlightstoallowamoreoverall optimaltrafficflowforthecityasawhole. Afurtherexampleofhowthiscanbeusedwithinsmart citieswouldberelatedtoreducingtheconcentrationsofairpollutionwithinthecity. Thiswouldbe achievedbyredirectingheavilypollutingvehiclesawayfromareaswithhighpollutionlevels. Asaresult,therehasbeenasignificantbodyofresearchdealingwiththealgorithmstoreduce traffic jams. The Dijkstra [2] and the A* algorithms [3] are the two most common path planning algorithms. Otherstudiesconcentrateontheintegrationofswarmintelligencealgorithms,suchas artificialantcolonyalgorithms[4],geneticandsimulatedannealingalgorithms[5]. However,most oftherecentlydevelopedalgorithmsaregenerallydesignedforstaticgraphs. Theyarenotdesigned to be employed for dynamic path planning in a real-time traffic environment. This is due to the unpredictabletrafficconditionsthatmightoccurinthedynamicenvironment. Recently,hybridsystemshavebeensuggestedtointegratealgorithmsineachgrouptocreate avigorousintegratedsolutiontoovercomeanydrawbacksorobstructionsoneachsingleapproach; forexample,thehybridgeneticalgorithmusingDijkstra’sheuristicmulti-objectiveoptimizationfor dynamicrouteplanningwiththepredictedtrafficinareal-worldroadnetwork[6]. Thereisasignificantincreaseindevelopingefficientsolutionsforimprovedmobilityinintelligent transportationsystems(ITS)[7]. Findingtheoptimalnavigationroute,fromasourcetoadestination, withinareasonabletimeisthekeytask. Differentcostfunctionscanbeapplied, suchasattaining theminimumtraveldistance(TD),minimumtraveltime(TT)orminimumfuelconsumption. ITSis comprisedofabroadrangeofwirelesscommunication-basedinformation,controlandtechnologies. When combined with the transportation system infrastructure and in vehicles themselves, this technologyhelpstomanagetrafficflow,reducetrafficjams,enhanceroadsafety,providealternative routestocommuters,reducefuelconsumption,time,moneyandvehicleemissionsincongestedurban areas[8]. AnefficientapproachcanbedevelopedwiththeaidoftheInternetofThings(IoT).TheIoTis anenvisagedcommunicationsystemthatallowsvariousdevices,suchaswirelesssensornetworks (WSN),mobilephones,mobilephonemasts,actuatorsandnear-fieldcommunicationdevices(NFC), to communicate and collaborate with each other to complete a common goal [9]. The Internet of Vehicles (IoV), a natural offshoot of the IoT, foresees all future vehicles to be connected, sharing informationrelatedtosafety,convenienceandinfotainment. Vehicletovehicle(V2V)andvehicleto infrastructure(V2I)communicationsystemsaretwosubsetofIoV.TheV2Vcommunicationsystem allowstwovehiclestointeractdirectlywithoutrelyingonfixedinfrastructure. Onthecontrary,theV2I communicationsystemallowsvehiclestocommunicatewithroadsideunits(RSU),suchassensors, traffic light controllers, intelligent signals, mobile phone masts, etc., which are installed as part of thesmartcityconcept[10]. V2VandV2Ihavebeenwidelyutilizedtoresolvevariouscomplications relatedtotransportationinsmartcities,inwhichthemostsignificantproblemisthetrafficcongestion. IoVsystemscansignificantlyimprovetrafficsafetyandconvenienceinsmartcities,byproviding driverswithtimelyinformationaboutroadconditionsandtravelsituations. Existingsolutionstothe trafficcongestionproblembyutilizingvehicularcommunicationrelyeitheronV2Vcommunication aloneorbothV2VandV2Icommunications. However,therearestillsomeLimitations,whichcanbe summarizedasfollows: 1. A large portion of the previous or current vehicle routing algorithms attempt to identify the minimumTDorTT.Generally,theycannotattainanactivetrade-off. 2. Utilizing just individual traffic information or a single cost function for the vehicle routing problemisnotsatisfactory. Differentnavigationcriteriashouldbeconsideredtofindtheoptimal pathofthedriver. Thiswillhelpdriverstohavedifferentnavigationoptions,whichcanbethe fastestroute,theleastcongested,theleastfuelconsumptionandtheleastairpollution. Sensors2016,16,1013 3of23 Thispaperpresentsandevaluatesanewmulti-objectiveimprovedsimulatedannealingtechnique fororderpreferencebysimilaritytotheidealsolution(ISATOPSIS)algorithmforcongestionavoidance basedonanIoVcommunicationsystem. ThemainobjectiveinISATOPSISistoprovidevariousroute decisionsaccordingtodifferentobjectivesinordertomeetthediversenavigationrequirementsof drivers;forexample,theminimumTT,TD,fuelconsumptionoratrade-offofallconditions. Inthis paper,twootheralgorithmshavebeenimplemented: simulatedannealingweightedsum(SAWS)and thesimulatedannealingtechniquefororderpreferencebysimilaritytotheidealsolution(SATOPSIS) forcompressionpurposes. ThecostfunctionofSAWShasbeenformulatedusingtheweightedsum method. ThecostfunctionofSATOPSISandISATOPSIShasbeenformulatedusingthemulti-attribute decisionmaking(MADM)method,whichiscalledthetechniquefororderpreferencebysimilarityto theidealsolution(TOPSIS)method[11]. TheresultsoftheproposedalgorithmISATOPSIShavebeen comparedtotheshortestpathDijkstraalgorithm(DA),SAWSandSATOPSIS. Theremainderofthepaperisorganizedasfollows: InSection2,aliteraturereviewofrelated workispresented. ThedetailsofSAWS,SATOPSISandISATOPSISaregiveninSection3. InSection4, aperformanceevaluationisprovided. Finally,conclusionsaredrawninSection5. 2. RelatedWork Numerous studies have been published on IoV as part of IoT to find a solution to the traffic congestionprobleminordertoreducetheTD,TTandfuelconsumption. Inthissection,adiscussion andreviewoftheseroutingalgorithmsisgiven. In [12], the algorithm aims to send geo-broadcast messages in V2V and V2I communications. The algorithm used an event-driven message instead of a beacon message sent periodically. Thismessageistransmittedtoallvehiclesthatarewithinaspecificrangetoprovidethedriverwith informationabouttrafficconditions. However,thisapproachfocusesonlyonprovidinginformationto othervehicleswithoutanymechanismtoavoidthecongestedarea. In[13],theauthorsemployV2Vcommunicationwithmessagesperiodicallysentbetweenvehicles. Theyproposeatechniquecalledcontent-orientedcommunication(COC)tosolvetrafficcongestion. Withthismethod,thevehicleevaluatestheroadtrafficdensityfrommessagesreceivedfromother vehiclesandperiodicallysendsthesedatatothefurthestvehicleswithincommunicationrange. Asa result,vehiclesobtaintrafficdensityinformationfordifferentlocationsanddetectcongestionlevelsby comparingtheroadtrafficdensitycalculatedwithaveragetrafficdensityvaluesfortheroadsegment. However,thisworkdidnotproposeamechanismtoavoidtrafficjams. In [14], a new Dijkstra algorithm was proposed. This algorithm encompasses new inputs for vehicle path planning, along with a reactive planner to update the selected path according to the requiredroutechanges. Ageneralclassificationofvehicleroutingalgorithmsandevaluationmetrics insmartcitiesisalsopresented. However,thisisdonewithoutanyperformanceevaluation. In [15], the authors evaluate four algorithms: static Dijkstra, static A*, dynamic Dijkstra and dynamicA*. Theevaluatedalgorithmshavebeenappliedinthreedifferenttestscenarios(citycentre, suburbanandrural). Accordingtotheperformanceevaluation,theauthorsrecommendtheA*-based algorithmsasthebestroutingalgorithmsinthedifferingroadenvironments. However,thedrawback ofthesealgorithmsisthattheyutilizeonlyasingleobjectivewhenfindingthealternativepaths,which leadstothetransferofthecongestiontootherroads. In[16],theauthorsproposeanalgorithmcalledcooperativetrafficcongestiondetection(CoTEC), whichusesfuzzylogiccontrolwithtwoinputs(thespeedandthedensityofvehicles)andoneoutput (congestionlevel). Thisalgorithmusescooperativeawarenessmessages(CAM)orbeaconmessages, sentperiodicallytoothervehiclestellingthemtoavoidthecongestedarea. Inaddition,itdetectsthe congestionbyusinganexternalmetriccalledthelevelofservice(LOS)developedbySkycomp. This involvestheclassificationofcongestionlevelsfromaerialsurveysofhighways. Onceobtained,this informationisthencombinedwithdatafromorganizations,suchaslocalauthorities. Asaresult,the performanceofthealgorithmisheavilydependentontheaccuracyofthedatafromtheseexternal Sensors2016,16,1013 4of23 organizations. Additionally, no mechanism is provided for drivers to find the optimal route for theirjourney. In[17],acentralizedframeworkisproposedtoobtainreal-timecoordinatesofvehiclesandtheir directionandvelocity,whichcanbeusedtodeterminetrafficcongestionlevels. Whencongestionis detected,vehiclesclosebyarere-routedusingtwoalgorithms. Thefirstalgorithm,dynamicshortest path(DSP),allocatestoeachvehicletheshortestpath(smallestTT)toreachtheirdestination. However, the drawback of the first algorithm is the probability of transferring the congestion to other areas. Therefore,thesecondalgorithm,randommultipathKshortestpaths(RKSP),determinestheKshortest paths for each vehicle and randomly allocates the vehicle to one of them. As a result of using K alternativepaths,itispossibletoensurethatthecongestionisnotjusttransferredtoadifferentroad. However, thedrawbackofthisalgorithmisthatthechoiceofroutesisarbitraryandusesasingle objectivefunction. In [18], the authors propose an ant-based congestion avoidance system (AVCAS). AVCAS integratestheaveragetravelspeedpredictionwithsegmentationofacitymaptodetectandreduce thecongestion. ItcollectsthetrafficinformationfromV2VcommunicationsandRSUtoforecastthe averagespeedoftheroads. Then,itusesaweightedsummethodtoformulatethecostfunctionforthe roads,wheretheweightshavebeencalculatedbytrialanderror. However,thisalgorithmstillchooses theshortestpathswhenre-routingthevehiclestoavoidthecongestion. Asaresult,thistransfersthe congestiontoanewroad. In[19],theauthorsdescribeadistributedreal-timeV2Vcongestionevasiontechnique.Congestion levelsaredetectedandvehiclesre-routedusingcongestionmessagescalledrequest/receivemessages (insteadofbeaconmessages). Whenavehiclereachesanintersectionandcanselectfromdifferent paths,itsendsacongestionrequestpackettoallneighbouringvehicleswithinitscommunicationrange toobtainthetrafficstatusinformation. Thismessageencompassesalistofroadsthatformpotential pathstotheirdestination. Thevehiclewillselectitsnextpotentialroutebasedonthereceivedtraffic data. Acongestionresponsemessageistransmittedonlywhenacongestionrequestisreceived. Ifthe congestionrequestmessageisreceivedtwice,thevehiclewillnotsendaresponsemessage,andthe congestionmessagewillbediscardedfromthedatabaseofthereceivingvehicle. Thisensuresthatno redundantinformationistransmitted. Otherwise,thevehiclewillbroadcasttheresponsemessage, which includes data about the congested roads available at the receiver. The vehicle that receives the congestion message request will select the road that has the minimum travel time. However, thismethodhastwodrawbacks. Firstly,itincludesahighdelaytimeduetosendingthetrafficdata. Secondly,theauthorshavenotconsideredthehighprobabilityandconsequencesoflostpacketsina builtuparea. In[20],theauthorssuggestanITSbasedonRSUsinordertoavoidtrafficcongestion. Thissystem istriggeredonlywhenanaccidentoccurs. Thevehiclesinvolvedstarttotransmitalertmessageswith thelocationoftheaccidenttotheothervehicleswithintheirtransmissionrange. Everyvehiclethat hasreceivedanalertmessagechecksifitisnearanaffectedarea. Ifitisfarfromtheaccidentpoint,the vehiclewilldeletethereceivedmessage. However,ifthevehicleisneartheaccidentpoint,itchecks itsroutetoestablishwhetheritisaffectedbytheaccidentornot. Atthisstage,anewroutecanbe calculatedandassignedtothevehicle. However,thedrawbackofthismethodisthatthere-routingis donebasedoneitherastaticDAorA*algorithm. Thiswouldtransferthecongestionfromoneareato another. Inaddition,thissystemdoesnotconsiderthetrafficcongestionthatoccursduetothehigh vehicledensitiesinagivenarea. Recently, in [21,22], the authors proposed two systems called geographical accident aware of reducing urban congestion (GARUDA) and a solution using cooperative re-routing to prevent congestionandimprovetrafficcondition(SCORPION)basedontheITSandIoV.TheGARUDAisa decentralizedsystem,anditoffersanewpathtothedriversconsideringallcarsandavailableroutes intheregionnearesttoanaccident. TheSCORPIONisacentralizedsystemthatcollectsthetraffic data from RSU, and it uses the concept of fuzzy logic to predict congested roads. However, both Sensors2016,16,1013 5of23 systemsproposethere-routingbasedonstaticDijkstraorA*algorithmsand,thus,willhavethesame shortcomingasin[15],i.e.,thepossibilityofmovingcongestiontoanewarea. Otherstudieshavereviewedthemostrelevantalgorithmstocalculatetherouteinthevehicle route problem (VRP), such as in [2–4,6] and [23]. In [23], a modified version of ACO is proposed inordertoreducethetraveltimeforvehiclesonthemove. Themodifiedantcolonyoptimization (MACO) is a variation of the classical ACO in which the idea of an ant colony has been reversed. Insteadofattractingthevehiclestowardroadsthathavehighpheromonelevels,theMACOalgorithm routesanddispersesthetraffictowardpathswithlowerpheromonevaluestoavoidthecongestion. However,thecostfunctionofthisalgorithmisverysimilartotheclassicalDijkstraalgorithm. Hence,thenoveltyofthisworkisthedevelopeddynamicmulti-objectiveoptimizationalgorithm, whichcombinessimulatedannealing(SA)withtheMADMTOPSIScostfunctiontoprovidethedriver with optimal paths. In addition, the proposed algorithm utilizes real-time data using IoV, unlike otherworks,wherethestaticalDAorA*isusedtore-routethevehiclesandleadstothetransferof congestionontootherroads. OurproposedapproachisalsobetterthantheACO-basedmethodsas theACOcanbecomestuckinlocaloptimasolutions(ratherthantheglobalsolution). Thisisbecause oftheACOupdatingthepheromonebasedonthecurrentoptimalroute[24]. Theproposedalgorithmhasfourfeatures: 1. ISATOPSISallowstransitionfromagoodsolutiontoaworsesolutionunderastrictcondition. Thisallowsthealgorithmtofindtheglobaloptimalsolutionandavoidbecomingstuckinlocal optimalsolutions. 2. ISATOPSIScanworkfordynamicpathplanningbycollectingreal-timetrafficdatafromIoVand efficientlyfindingalternativeroutesforthedriver. 3. ISATOPSIScanoptimizemorethanonecriteriausingtheMADMTOPSISmethod,whichallows alternativeroutestobejudgedondifferentcriteria. 4. ISATOPSIS periodically detects and avoids congestion by selecting the paths that have the minimum traffic, CO emissions, fuel consumption, as well as travel time. This is due to 2 combiningdifferentnavigationattributesinthecostfunction. 3. SystemDescription IoVconsistsoftwocomponents: V2VandV2Icommunication. TheV2Vsystemisanon-board WSN,whichisinstalledinthevehiclesthemselves. Thesensorsallowthevehiclestosendandreceive information,suchasspeed,locationanddirection[10]. TheV2Isystemisaroadsideunitnetwork comprisingsomeoftheinfrastructurerelatedtosmartcities,e.g.,trafficsensorsdeployedalongthe roadsandatintersections. Inthispaper,bothsystemshavebeenusedwiththe“hello”protocol[25]orbeaconmessages, whichfocusontacklingtheproblemofmonitoring,trafficandreducingcongestion. Figure1showsthe V2VandV2Iarchitectures. Inthisprotocol,eachvehiclewillhaveanoverviewoftheaveragespeed anddensityofvehiclesontheroadnetwork. Thisallowsthemtochoosetheoptimalpathtoreachtheir destination. Inthissection,wedescribeourproposedsystembyspecifyingthedatadissemination methodology,theroadnetworkmodel,simulatedannealingofSAWSandSATOPSISandanimproved simulatedannealingTOPSISalgorithm. 3.1. DataDissemination Thedatahave beentransmittedusingbeacon messages, andthe proposedprotocolworksas follows: thevehiclestransmittheiraveragevelocityand“roadId”totheirneighbouringRSUsthrough beaconmessages. EachRSUholdsadatastructurecontainingtheaveragespeeds,roadIdsandroads lengthofallvehicleswithinitstransmissionrange. Theaveragespeedsarefoundfromthespeed measurementsoverthepreviousfiveseconds(onemeasurementasecond). Iftheaveragespeedis lessthanorequaltoavelocitythresholdpredeterminedbythedesigner,thecongestiondetectionis initiated,inwhichcongestedroadsareidentified. TheRSUwillthenverifywhetherornottobroadcast Sensors2016,16,1013 6of23 thedatafromreceivingbeaconmessages. ThedatawillbeusediftheRSUdoesnotreceiveaduplicate messagewiththesameroadId. WhenevertheRSUreceivesanewbeacon,itupdatesitsdatastructure andbroadcaststhedatatovehicleswithinitstransmissionrange. Asaresult,congestedroadscanbe excludedfromthemapandanewroutecalculatedattheapplicationlayerofV2Vcommunication usingtheISATOPSISalgorithm. Ro ad Side Unit (RSU) Vehicle on-board sensors Figure1.InternetofVehicles(IoV)roadnetworkinfrastructure. 3.2. RoadNetwork The road network can be modelled as a directed graph G = (N,E), where N corresponds to theintersections(nodes)andE = {e ,e ,...,e}correspondstotheroadsegments(edges). Theroad 1 2 i matrix Acanbeformulatedasfollows: supposee achintersectioncontainsnroads;eachoftheroads containsthejattributevalueintheroadnetwork: C C L S A r r 1 11 12 A r r 2 21 22 . . . A = .. .. .. (1) A r r n n1 n2 w w 1 2 Thenormalizedroadmatrixhasbeenobtainedusingthefollowingequation: x kj rkj = (cid:115) where k =1,...,n; j =1,2 (2) n ∑(x )2 kj k=1 wherer = {r |k = 1,...,n; j = 1,2} arethenormalizedperformancevaluesofeachC andC , kj L S respectively. X = {x |k =1,...,n; j =1,2}denotesthesetofperformancevaluesofeachC andC , kj L S respectively. w = {w | j =1,2}denotesthesetofweights;V = {v ,v ,...,v }isthesetofvehicles; j 1 2 j Sensors2016,16,1013 7of23 andA = {A |k =1,...,n}arethealternativeroadsforeachvehicleinV.Everyvehiclev periodically k j sends a message msg thatcontains {roadId ,averagespeed ,position ,route ,destination } to the j j j j j j neighbouringRSUs. Twoparametershavebeenusedinouroptimization: 1. Road length C = {r |k = 1,...,n; j = 1} represents the normalized length in a directed L kj graphGforeachalternativein A. 2. AveragevelocityC = {r |k = 1,...,n; j = 2}representsthenormalizedaveragespeedof S kj eachvehicleatacertainperiodin A. 3.3. SimulatedAnnealingofSAWSandSATOPSIS SAisanoptimizationapproachfirstproposedby[26]thatimitatestheprocessofannealingin mineralogy. Itallowstransferencefromagivensolutiontoaworsesolutionunderstrictconditions. Thisallowsthealgorithmamutationtomovefromlocalminimaormaximatowardstheglobaloptima. SAbeginswithasetofrandomly-chosensolutions. Eachiterationproducesanewsolutionofthe statevector,whichisfoundbasedonthecostfunction. Anewsolutionwithahighcostisaccepted immediately. However,asolutionwithalowercostthanthepreviousonecanbeacceptedwitha transitionprobability. Inthiswork, K randompathshavebeengeneratedfrommatrix A. Figure2 showshowthepathsaregenerated. Initial path P = {𝒓 , 𝒓 , ……, 𝒓 } 𝒔 𝟏 𝒅 Randomly select a road 𝒓 from road matrix A 𝒊 No Is the path P is feasible Yes Delete the 𝒓 from A 𝒊 Is the No destination reached? Yes Return P and repeat the process until the desired number of paths are achieved Figure2.Theprocedureofgeneratingarandompath. SAhasbeenusedwithtwodifferentcostfunctions(theweightedsumandTOPSISmethods)to selecttheoptimalpathfromtheKrandompaths. Everyv inV hasasetofalternativepathsfromR , j k Sensors2016,16,1013 8of23 whereR isamatrixofKrandompathsgeneratedfrom A. EverypathinR containsthenumberof k k roads A ,andthecostfunctionofeachpathhasbeencomputedasthesumofthecostsofallofthe k roadsitcontains. Algorithm1describestheprocedureofchoosingapathfromKrandompaths,whereX isthe c current solution, which is generated randomly from A. T is the temperature parameter, which is moderately decremented with time. The constant α is the cooling rate used to gradually decrease thevalueofT. Whenthetemperatureparameterhasaveryhighvalue,i.e.,T → ∞,anewpathX n isselectedrandomlyfrom R . Thecostfunctionisevaluatedfor X (N(X ))andcomparedtothe k n n previousvalueofthecostfunction(C(X )). Ifthecostfunctionishigherthanthepreviousvalue,then c thesolutionisaccepted. Evenifthenewsolutionisnotappropriate(meaningthecostofthenew solutionislessthanorequaltothecurrentsolution),itisacceptedwithsomeacceptanceprobability. Thishelpstoexpandthesearchandavoidslocaloptima. WhenTapproacheszero,pathswithahigh costhaveahighprobabilityofbeingaccepted. The SAWS and SATOPSIS find up to three alternatives for vehicles that have the same source/destination. Whentheoptimalpathsarefound,theyareorderedbasedonthecostfunction. The vehicles that have the same source/destination are distributed on these three optimal paths. Thiswillensurethevehiclesaresharedbetweenthepathsandensuresthetrafficloadisbalancedon themap. Algorithm1Thesimulatedannealingalgorithmforenhancingmobility. 1: Xc = Xc0 Initialrandomsolution 2: T = T0Aninitialtemperature 3: α=Coolingrate 4: sb =Currentbestsolution 5: sb ←Xc 6: WhileT > Tm whereTm istheminimumtemperature 7: GeneratearandomneighboursolutionXn fromRk 8: IfN(Xn)>C(Xc) 9: MovetoXn 10: Acceptchangesb ←Xn 11: ElseIfN(Xn)≤C(Xc)Then 12: MovetoXn withtransitionprobability 13: Pt =1/1+exp(C(Xc),N(Xn),T) 14: Endif 15: T = αT 16: Endwhile(ifT < Tm) 17: Returnsb 3.4. AnImprovedSimulatedAnnealingTOPSISAlgorithm In this section, SA has been implemented for dynamic path planning using the TOPSIS cost functionmethod. Thedecision-makingprocessoftheproposedapproachisconstructedasfollows: 3.4.1. Off-LineComputationofPathPlanning SAbeginswithoff-lineroutecalculation,inwhicheveryv inV hasasetofroadsfrom Aand j everyroad A in AhasacostfunctionformulatedusingTOPSISmethod. Thealgorithmdescriptionof k theSAisoutlinedinAlgorithm1,exceptStep1,whereX isthecurrentsearchsolutionoraninitial c feasiblepath,whichisanoptimalpathfrom R . Whenthetemperatureparameterhasaveryhigh k valuethenewsearchsolution,X inStep7isconstructedbasedonX asfollows: n c Sensors2016,16,1013 9of23 1. An initial optimal path Xc = {rs,r1,...,ri,ri+1,...,rl−1,rl,rd} where ri means the i-th roadsegment. 2. Theperturbation(seeFigure3)consistsofthefollowingthreesteps. (a) Tworoadsr andr ,calledbaseroads,arechosenrandomlyintheX path. i l c (b) Apathisconstructed,usingr asanoriginandr asadestination. i l (c) ThepathXc = {rs,r1,...,ri,ri+1,...,rl−1,rl,rd}isreplacedby(ri,ri+´1,...,rl−´1)togivea newpathXn = {rs,r1,...,ri,ri+´1,...,rl−´1,rl,rd}. 3. Checkthefeasibilityofthenewpath. 4. Ifitsnotfeasible,thenrepeattheprocess. Otherwise,useSAasinAlgorithm1andcomparethe costofthenewpathtothepreviouspath. 𝒓𝒊 𝒓𝒊′+𝟏 ..…….… 𝒓𝒍′−𝟏 𝒓𝒍 𝒓𝒔 𝒓𝟏 ..…… 𝒓𝒊 𝒓𝒊+𝟏 ..…….… 𝒓𝒍−𝟏 𝒓𝒍 ..…… 𝒓𝒅 Figure3.TheprocedureforconstructinganewpathXnbasedonaninitialpathXc. 3.4.2. On-LineComputationofPathPlanning Start Journey Initial optimal path from source to destination Drive on current path Reach an Intersection Is there an Yes update? No Does this Follow the same No update effect path the current path? Yes A. Disclose the congested Is the roads No destination B. Discard the congested reached? roads from the map C. Calculate a new optimal route using ISATOPSIS Yes End Journey Figure4.Flowchartofthesimulatedannealing(SA)congestionavoidancemechanism. Asdiscussedintheoff-linecomputationsection,thevehiclesusetheroutegeneratedbyoff-line pathplanningtotravelthroughthecity;theon-linepathplanningisthentriggeredtoautomatically computeanalternativeroutewhencongestionisdetectedasfollows: Whenthevehiclereachesan intersection and enters the RSU transmission range, it will receive updated data with the table of congestedroads. Then,thevehicleevaluatesitscurrentroute. Iftheevaluationshowsthatthecurrent Sensors2016,16,1013 10of23 routehasnotbeenaffectedbytheupdateddata,thevehiclewillkeeptravellingalongthecurrent route. Otherwise,ifthevehicleislikelytoenteracongestedroad,SAwillbeactivatedandreloaded withtheupdatedsearchspace. Theupdatedsearchspacecontainsthecurrentstatusofthevehicle, itscurrentlocationandthenewcostoftheroadsegments. Thealternativeroutewillbecomputed fromtheupdatedsearchspacetoallowthevehicletotravelfromitscurrentlocationtothedestination. Figure4showstheprocedureofthecongestionavoidanceusingISATOPSIS. 3.5. CalculatetheWeightsofSAWS,SATOPSISandISATOPSIS Inthispaper,thestandarddeviation(SDV)methodisusedtodeterminetheweightsofmultiple objectives. TheweightintheMADMproblemreflectstherelativeimportanceofthevariousobjectives. Theweightsofthedifferentcriteriahavebeennormalizedtotransformthedifferentscalesandunits intocommonmeasurableunitsusing: r − min r kj kj (cid:48) 0≤j≤m R = (3) kj max r − min r kj kj 0≤j≤m 0≤j≤m (cid:48) (cid:48) whereM = (R )m×nisthematrixafterrangenormalizationandmaxrkjandminrkjarethemaximum andtheminimumvaluesofthecriterion(j)inC andC ,respectively. L S (cid:48) R ∈ [0,1] k =1,...,n; j =1,2 kj SDV is the standard deviation that is calculated independently for every j-th criterion using the j (cid:48) (cid:48) normalizedmatrix M = (R )m×n asshownin(4): (cid:115) SDV = 1 ∑n (R(cid:48) −R¯ )2 (4) j n kj j k=1 where: R¯ = 1 ∑n R(cid:48) (5) j n kj k=1 andR¯ isthemeanofthevaluesofthej-thcriterioninC andC afternormalizationandj =1,2. j L S FromEquation(4),theweight(w )ofthecriterion(j)inroadmatrix Acanbedefinedas: j SDV j w = (6) j 2 ∑ SDV j j=1 3.6. SimulatedAnnealingWeightedSumMethod A multi-objective problem is often solved by combining the multiple objectives into one single-objectivescalarfunction.Thisapproachisingeneralknownastheweighted-sumorscalarization method. Inthispaper,thecostfunctionofthesimulatedannealinghasbeenformulatedusingthe weightedsummethodasbelow: f =Max{w C +w C } (7a) 1 L 2 S n ∑ C = r (7b) L k1 k=1 n ∑ C = r (7c) S k2 k=1
Description: