动态公交车辆运行时间预测模型

25320106JOURNALOFSYSTEMSENGINEERINGVo.l25No.3Jun.2010姚宝珍1,杨成永1,于滨2(1.,100044;2.,116026):准确预测公交车运行时间是先进的出行者信息系统(ATIS)的核心.本文应用支持向量机(SVM)进行公交车辆的运行时间预测,其目的是要验证SVM在运行时间预测领域的可行性.为了调整不同阶段历史数据对预测结果的影响引入了衰减因子,并应用一种自适应算法来动态调整预测误差.然后以大连市23路公交车对该模型进行来了检验.结果显示,带有衰减因子和自适应算法的支持向量机算法具有很好的预测精度和动态性能.:;;;;:U121:A:1000-5781(2010)03-0365-06DynamicbustraveltimepredictionmodelYAOBaozhen1,YANGChengyong1,YUBin2(1.SchoolofCivilEngineering&Architecture,BeijingJiaotongUniversity,Beijing100044,China;2.SchoolofTransportaitionManagemen,tDalianMaritimeUniversity,Dalian116026,China)Abstract:Effectivepredictionofbusarrivaltimeisacoreforadvancedtravelerinformationsystem.Supportvectormachines(SVM)areappliedtopredictingbustraveltimes.TheobjectiveofthispaperistoexaminethefeasibilityandapplicabilityofSVMinthevehicletraveltimeforecastingarea.Adecayfactorisintroducedtoadjusttheweightsbetweennewandolddata.Also,anadaptivealgorithmisusedtoimprovethepredictionresults.TheSVMwiththedecayfactoristestedwiththedataofthenumber23busrouteinDaliancity.ResultsshowthattheSVMwiththedecayfactorandtheadaptivealgorithmhasbetterpredictionaccuracyanddynamicperformancethanotherexistingalgorithms.Keywords:prediction;traveltime;supportvectormachine;decayfactor;adaptivealgorithm0,..,.,:2008-03-03;:2009-03-27.:(50978020);(20050151007);(20070151013);(141065522).,,,,.[1-2]Kalman[3-5][6-7].(SVM)[8-9],,,.SVM,SVM,SVM.,,,,,[10].,,,,,.11.1SVM,,,SVM()().,SVM().,SVM,,.[8,11].l(x1,y1),(x2,y2),,(xl,yl)(xi!XRn,yi!YR),SVM,xH,,.,f(x)=(x)+b(1),Q.Q=12∀∀2+Cl#li=1L!(yi,f(xi))(2),1,,;2,,C0,.!-L!(yi,f(xi))(3).L!(yi,f(xi))=max(|yi-f(xi)|-!,0)(3)(2)2,aia*i,-#li=1(ai-a*i)xi=0(4),f(x)=#li=1(ai-a*i)(xi)∃(x)+b(5)K(xi,xj)(5)f(x)=#li=1(ai-a*i)K(xi,x)+b(6)K(xi,xj)xixj(xi)(xj),K(xi,xj)=(xi)∃(xj),.1.2(SVMDF),,,..,,,,%&,.,,%&%&.,,.,∋366∋25,t^hLhv^h(k)=f(x)(7)x=(vh(k-1),∀vh(k-2),,∀k-n-1vh(k-n)),t^hSVMDFkh;v^h(k)SVMDFkh;Lhh;vh(k-n)nh;∀,0∀1.∀.∀,%&,.∀%&.(7).∀,.,,(),.1.3,,1.2.,.,[6].gh(k)(0(gh(k)(1)SVMDF.(gh(k))gh(k),,.th(k)kh,t^h(k)*kh.t^h(k)*=(1-gh(k))t^h(k)+gh(k)#h(k-1)*(8)gh(k)=∃h(k-1)∃h(k-1)+∃h(k-1)*/gh(k-1)(9)#h(k-1)*=th(k-1)-t^h(k-1)*(10)#h(k-1)=th(k-1)-t^h(k-1)(11)∃h(k-1)*=E((#h(k-1)*)2)(12)∃h(k-1)=E((#h(k-1))2)(13)gh(k);#h(k-1),#h(k-1)*SVMDFk-1h;∃h(k-1),∃h(k-1)*SVMDF.2SVMDF,,,,,.,23.23,14.5km,19.,23.(2)(4)()),1.85km,1.1Fig.1Sketchschemeoftestroute2.1,2006910,(6∗30-7∗30)∋367∋3:(10∗00-11∗00)),520.SVM,,,.2.2.[3,11],(C,!,%).(2-2,2-5,1.22).,SVMDF.2(MAPE),MAPEMAPE=1J#Jj=1|th(j)-t^h(j)|th(j)+100%(14)J.2(a),,8~10,0.70~0.85,,.,n=8,∀=0.85.,,n=6,∀=0.8.2(a)(b),,,,,,,,2(b),6,.,,,;,%&.2.3,3,,,10%,20%,.SVMDF(SVMDFAA),,(SVM)(HMP).SVMSVMDF,t,h(k)=1m#mi=1th(i)(15)34.3,3,HMP(),,.SVMDFSVMVC,,.,3,SVMDFSVM,∋368∋25,.SVM,,,,SVMDFAA.,,,.34Fig.3Comparisonofpredictionperformancesoffourmodels3.SVM,,,,.,,.,23,,,.,,,.,.:[1]DeLurgioSA.ForecastingPrinciplesandApplications[M].NewYork:McGrawHil,l1998.[2],,.[J].,2007,33(21):281-282.ChenSikang,ZhanChengchu,ChenLianggu.iPredictionmethodofbusarrivaltimebasedonlinktraveltime[J].ComputerEngineering,2007,33(21):281-282.(inChinese)[3],,.SVMKalman[J].,2008,21(2):89-92.YuBin,YangZhongzhen,ZengQingcheng.BusarrivaltimepredictionmodelbasedonsupportvectormachineandKalmanfilter[J].ChinaJournalofHighwayandTransport,2008,21(2):89-92.(inChinese)[4],.[J].,1999,19(9):74-78.ZhuZhong,YangZhaosheng.ArealtimetraveltimeestimationmodelbasedontheKalmanfilteringtheoryfortrafficflowguidancesystem[J].SystemsEngineering:Theory&Policy,1999,19(9):74-78.(inChinese)[5],,.[J].:,2008,36(10):1355-1361.XuTiandong,SunLijun,HaoYuan.Realtimetrafficstateestimationandtraveltimepredictiononurbanexpressway[J].JournalofTongjiUniversity:NaturalScienceEdition,2008,36(10):1355-1361.(inChinese)[6]ChienIJ,DingY,WeiC.Dynamicbusarrivaltimepredictionwithartificialneuralnetworks[J].JournalofTransportationEngineering,AmericanSocietyofCivilEngineers,2002,128(5):429-38.[7]ChenM,LiuXB,XiaJX,eta.lAdynamicbusarrivaltimepredictionmodelbasedonAPCdata[J].ComputerAidedCivilandInfrastructureEngineering,2004,19(5):364-376.[8]VapnikVN.TheNatureofStatisticalLearningTheory[M].NewYork:SpringerVerlag,2000.[9].[J].,2006,21(4):410-413.TangWanme.iNewforecastingmodelbasedongreysupportvectormachine[J].JournalofSystemsEngineering,2006,21∋369∋3:(4):410-413.(inChinese)[10],,,.GARCHEWMA[J].,2006,38(19):1572-1575.LiuYifang,ChiGuota,iYuFangping,eta.lForecastmodeloffuturespricebasedonGARCHandEWMA[J].JournalofHarbinUniversityofTechnology,2006,38(19):1572-1575.(inChinese)[11],,.[J].,2007,27(4):160-165.YuBin,YangZhongzhen,LinJiany.iBusarrivaltimepredictionusingsupportvectormachines[J].SystemsEngineering:Theory&Practice,2007,27(4):160-165.(inChinese):(1976∋),,,,:,,Emai:[e-mail];(1966∋),,,,,:;(1977∋),,,,.:,.(297)[15]BarahonaM,PecoraLM.Synchronizationinsmallworldsystems[J].PhysicalReviewLetter,2002,89(5):054101(1-4).[16]YookSH,JeongH,BarabsiAL,eta.lWeightedevolvingnetworks[J].PhysicalReviewLetter,2001,86(5):5835-5838.[17]ZhengDF,TrimperS,ZhengB,eta.lWeightedscalefreenetworkswithstochasticweightassignments[J].PhysicalReviewE,2003,67(4):040102(1-4).[18]WangSJ,