Graph-based Modeling of Information Flow Evolution and Propagation under V2V Communications based Advanced Traveler Information Systems

Transcription

1 Graph-based Modelng of nformaon Flo Evoluon and Propagaon under V2V Communcaons based Advanced Traveler nformaon Ssems Graph-based Modelng of nformaon Flo Evoluon and Propagaon under V2V Communcaons based Advanced Traveler nformaon Ssems 1 Yong Hoon Km & Srnvas Peea* Purdue Unvers, 550 Sadum Mall Drve, Wes Lafaee, N 47907, USA Absrac: A vehcle equpped h a vehcle-o-vehcle (V2V) communcaons capabl can connuousl updae s knoledge on raffc condons usng s on eperence and anonmousl obaned ravel eperence daa from oher such equpped vehcles hou an cenral coordnaon. n such a V2V communcaons based advanced raveler nformaon ssem (ATS), he dnamcs of raffc flo and ner-vehcle communcaon lead o he me-dependen vehcle knoledge on he raffc neork condons. n hs cone, hs sud proposes a graphbased mul-laer neork frameork o model he V2Vbased ATS as a comple ssem hch s comprsed of hree coupled neork laers: a phscal raffc flo neork, and vrual ner-vehcle communcaon and nformaon flo neorks. To deermne he occurrence of V2V communcaon, he ner-vehcle communcaon laer s frs consruced usng he me-dependen locaons of vehcles n he raffc flo laer and ner-vehcle communcaon relaed consrans. Then an nformaon flo neork s consruced based on evens n he raffc and ner-vehcle communcaon neorks. The graph srucure of hs nformaon flo neork enables he effcen rackng of he me-dependen vehcle knoledge of he raffc neork condons usng a smple graphbased reverse search algorhm and he sorage of he nformaon flo neork as a sngle graph daabase. Furher, he proposed frameork provdes a rerospecve modelng capabl o arculae eplcl ho nformaon flo evolves and propagaes. These capables are crcal o develop sraeges for he rapd flo of useful nformaon and raffc roung o enhance neork performance. also serves as a basc buldng block for he desgn of V2V-based roue gudance sraeges o manage raffc condons n congesed neorks. Snhec epermens are used o compare he graph-based approach o a smulaon-based approach, and llusrae boh memor usage and compuaonal me effcences. 1 NTRODUCTON Communcaons echnologes enable echnologcal advances o be negraed no he ransporaon ssem and vehcles o foser objecves such as congeson mgaon, safe mprovemen, and raffc neork performance enhancemen. n hs cone, vehcle-o-vehcle (V2V) communcaons can be leveraged n an advanced raveler nformaon ssem (ATS) o allo vehcles o accumulae her on ravel eperence daa and communcae h oher vehcles hn communcaon range o echange ravel eperence daa hou an cenral coordnaon. Hence, V2V communcaons capables n an ATS can provde a daa-rch envronmen for ravelers based on nformaon ransmed anonmousl from vehcles hou he requremen of addonal nfrasrucure. Thereb, can poenall provde an enhanced range of aareness of raffc condons o ravelers. The dnamcs of vehcular raffc flo, ner-vehcle communcaon, and raffc nformaon flo are he hree underlng facors ha shape a V2V-based ATS. A comple characersc of hs ssem s ha hese facors hemselves nerac h each oher. Due o hese neracons, a V2V-based ATS can be veed as conssng of coupled laers nvolvng raffc flo, nervehcle communcaon, and nformaon flo, n hch evens n he dfferen laers are nerdependen. n hs sud, e use he erm nformaon flo o denoe he flo of he nformaon on he me-dependen lnk ravel me eperenced b a vehcle, and refer as a un of nformaon. Ths nformaon s no processed for congeson or an ncden deecon (Yang and Recker, 2008), and/or oher applcaons hrough daa fuson/updae. The analss of he propagaon of a sngle un of descrpve nformaon has been proposed b varous analcal approaches (Wu e al., 2005; Wang, 2007; Km e al., 2014). B conras, as vehcles generae her on lnk ravel me eperence daa over me and space, a V2Vbased ATS enals he propagaon of mulple uns of

2 2 Km and Peea nformaon. The se of ravel eperence daa on an equpped vehcle, based on on eperence or obaned hrough ner-vehcle communcaon, s referred o as he vehcle knoledge. Due o he hghl decenraled naure of he V2V-based ATS n he absence of cenraled coordnaon of nformaon provson, and he dnamcs assocaed h he raffc and nformaon flos, dfferen vehcles ma have dfferen me-dependen knoledge of he neork raffc condons. n hs cone, hs sud focuses on modelng he nformaon flo evoluon and propagaon ha lead o he dnamcs of vehcle knoledge n V2V-based ATS as a buldng block o develop coordnaed nformaon provson sraeges ha addonall ould requre an undersandng of ho he vehcle knoledge ould affec he drver acons. Ths s because he esmaed neork raffc condons based on a vehcle s me-dependen knoledge can be used b s drver for roue choce decsons. The roue choce decsons of V2V-equpped vehcles ould hen lead o he raffc neork flo evoluon and nfluence he dnamcs of he nformaon flo due o her neracons. To reerae, hs sud does no nend o deermne he drver roue choces based on he nformaon conen n he me-dependen vehcle knoledge. Thus, he denfcaon of he me-dependen vehcle knoledge addressed here s a subproblem of he broader V2V-based ATS ha seeks o address user/ssem objecves n congesed raffc neorks, possbl n coordnaed conrol sengs. Ths aspec s renforced furher hen dscussng he concepual frameork n Fg.1. Several sudes (L e al., 2001; Gupa and Kumar, 2000; Vuuru and Oguch, 2007) orgnang from he communcaons doman prmarl focus on ho he vehcle dnamcs affec he ner-vehcle communcaon effecveness beeen vehcles; for eample, a dfferen speeds. Hence, hese sudes have sough o address he negraon of he dnamcs of he raffc flo and he nervehcle communcaon. Oher sudes (Wu e al., 2005; Fgbbons e al., 2004; Schroh e al., 2006) propose frameorks ha ncorporae a raffc flo smulaor (such as Paramcs and CORSM) and a reless (ner-vehcle communcaon) neork smulaor (such as NS-2 and Qualne, or a smple analcal model) o derve some descrpve nsghs on he neracons beeen he raffc flo movemen and he ner-vehcle communcaon. Hoever, he aforemenoned sudes focus prmarl on he feasbl and he relabl of he V2V communcaon ssem for praccal applcaons, and do no eplcl address he modelng of he dnamcs of he raffc nformaon flo n he V2V-based ATS cone. Thereb, he address onl o of he hree coupled laers denfed hereofore. Movaed b he need o anale he dnamcs of he neracons among he raffc flo, ner-vehcle communcaon, and nformaon flo a he neork level, recen smulaon-based sudes (Echler e al., 2005; Wu e al., 2005; Km, 2010; Schmd-Esenlohr e al., 2007) n he ransporaon doman seek o esmae he dnamc raffc condons hrough smple daa updae mechansms so ha ravelers can use hs nformaon o make roung decsons under V2V communcaon ssems. The ncorporae a mcroscopc raffc flo model and a se of ner-vehcle communcaon consrans o deermne vehcle knoledge. Thereb, hese smulaon-based approaches enal a descrpve capabl o denf he me-dependen knoledge of each vehcle n erms of her on ravel eperence daa and such daa obaned from oher vehcles hrough V2V communcaon. Hoever, hese approaches do no have a rerospecve capabl o arculae eplcl ho nformaon flo evolves and propagaes, parcularl n erms of s lnkage o he neracons h he raffc flo and ner-vehcle communcaon dnamcs. Tha s, he canno rack hen and from hom a specfc un of V2V communcaon-based ravel eperence daa locaed n a ceran vehcle s knoledge reaches, and hen and o hom propagaes from. Such spaoemporal capables are crcal o develop sraeges for boh he rapd flo of useful nformaon and raffc roung o enhance neork performance. The lack an eplc model for he nformaon flo laer a he neork level n smulaon-based and analcal approaches precludes an undersandng of he fundamenal relaonshps beeen he dnamc neracons among he hree laers, and he evoluon of equpped vehcles knoledge n me and space. Ths s crcal for hree realorld objecves: () o denf and/or desgn nformaon flo sraeges/paradgms ha lead o he rapd propagaon of useful nformaon (n he sense of enhancng he raffc neork performance), () o develop argeed V2V-based roung sraeges o manage raffc neork condons, and () o desgn a V2V-based ATS so ha such communcaons are relable and successful. Ths paper seeks o fll hs ke gap n he leraure b proposng an negraed graph-based mul-laer neork frameork o model he V2V-based ATS as a comple ssem hch s comprsed of hree coupled neork laers: raffc flo neork, ner-vehcle communcaon neork, and nformaon flo neork. The proposed graph-based frameork provdes an eplc rerospecve modelng capabl o arculae ho nformaon flo evolves and propagaes beond he curren descrpve capabl afforded b smulaon-based approaches. To do so, he frameork s modeled as a se of neracng neorks n hch nformaon flo occurs as a resul of evens n he raffc flo and ner-vehcle communcaon neorks. These evens are he ravel eperence daa generaon n he raffc flo neork, and he V2V communcaon occurrences n he ner-vehcle communcaon neork. n parcular, he dnamcs of raffc flo are represened n erms of he spaoemporal

3 Graph-based Modelng of nformaon Flo Evoluon and Propagaon under V2V Communcaons based Advanced Traveler nformaon Ssems vehcle rajecores n he raffc neork, and he feasbl of ner-vehcle communcaon among vehcles s capured n he ner-vehcle communcaon neork usng consrans. Hence, o smulae he nformaon flo evoluon and propagaon ha characere he vehcle knoledge n a V2V raffc ssem, he follong are assumed o be knon: () ravel eperence daa from he phscal raffc neork, hch represens he acual ravel eperences of vehcles (n erms of he me a vehcle eners a lnk and s eperenced ravel me on ha lnk) along her roue rajecores, and () he ner-vehcle communcaon consrans arsng from he communcaons neork echnolog (n erms of he communcaon range, nerference and banddh). Based on he knon enes n he V2V-based ATS, he sud frs consrucs he vrual ner-vehcle communcaon laer ha llusraes ner-vehcle communcaon evens deermned b he knon locaons of he vehcles and he echnologcal consrans. Then, consrucs he nformaon flo neork laer based on he evens n he oher o laers. Fnall, seeks o denf he vehcle knoledge of all vehcles n space and me o eplan ha nformaon s obaned b each equpped vehcle. Through hese modelng processes, he graph-based approach provdes a rerospecve capabl o eplcl llusrae ho nformaon flo evolves and propagaes, parcularl n erms of he lnkage o he neracons h he raffc flo and ner-vehcle communcaon dnamcs. The proposed nformaon flo neork has o ke modelng characerscs. Frs, has a graph srucure ha llusraes he nformaon flo evoluon and propagaon. Ths enables he effcen use of a graph-based search algorhm o oban a vehcle s ravel eperence daa based on s raversng a conneced subgraph of he nformaon flo neork. Second, sores daa usng an effcen graph daabase. Graph daabases (Robnson e al., 2013) use he graph as a daa srucure ha s opmed for he effcen sorng and processng of dense, nerrelaed daases. Akn o oher sudes ha nvolve graph daabases such as socal, caon, and bologcal neorks, an nformaon flo neork emplos a graph daabase b assemblng nodes and lnks no a sngle graph srucure. Thus, he nformaon flo neork elmnaes daa represenaon redundances so ha he same pece of daa s no sored n more han one place, and nformaon flo evoluon and propagaon s represened usng dreced lnks. Also, snce a graph-based search algorhm o characere n he nformaon flo neork performs a local search and s no concerned h he neork se, adapng a graph daabase enables he developmen of compuaonall effcen soluon mehodologes. n summar, hs sud has o prmar objecves n he V2V-based ATS cone: (1) o model he nformaon flo evoluon and propagaon hrough an negraed mullaer neork modelng frameork, and (2) o characere he spaoemporal evoluon and propagaon of nformaon flo hrough a graph srucure. As he ke sud conrbuon, he proposed frameork provdes a rerospecve modelng capabl o undersand he nformaon flo evoluon and propagaon eplcl b capurng he dnamcs of he neracons nvolvng he raffc flo and he ner-vehcle communcaon laers. Thereb, he proposed graph srucure of he nformaon flo neork can llusrae he nformaon flo evoluon and propagaon. Anoher conrbuon s he ssemac modelng of he raffc flo, ner-vehcle communcaon, and he nformaon flo neorks o capure her neracons n he V2V-based ATS cone. From a praccal sandpon, he proposed graph-based approach can effcenl model he V2V-based ATS b leveragng a sngle graph daabase. The remander of he paper s srucured as follos. Secon 2 presens he graph-based mul-laer modelng frameork. The srucural and funconal properes of he nformaon flo neork o denf he characerscs of nformaon flo evoluon and propagaon are presened hereafer. Secon 3 dscusses he rerospecve capables assocaed h modelng he nformaon flo evoluon and propagaon, and he epeced benefs of modelng he nformaon flo neork n erms of a graph srucures. Secon 4 descrbes he snhec epermens and dscusses he assocaed resuls and nsghs. Secon 5 presens some concludng commens. 2 PRELMNARES 2.1 The negraed mul-laer neork frameork As llusraed b Fg. 1, he V2V-based ATS can be veed as an negraed mul-laer neork frameork conssng of hree neork laers: raffc flo neork, ner-vehcle communcaon neork, and nformaon flo neork. Ther srucures are deermned based on he phscal raffc neork. Hence, he hree neork laers have neracons. For eample, he ner-vehcle communcaon neork evoluon s lnked o he raffc flo neork hrough dnamc vehcle rajecores and ner-vehcle communcaon consrans. The nformaon flo neork s dependen on he oher o laers hrough evens n hem; specfcall hrough he ravel eperence daa generaon n he raffc flo neork and he V2V communcaons n he ner-vehcle communcaon neork. Thereb, he nformaon flo evoluon and propagaon can be depced as a neork hose nodes correspond o evens ha occur n he raffc flo and ner-vehcle communcaon neorks, and hose lnks ndcae he drecon of nformaon flo propagaon. Ths faclaes he analss of her spaoemporal neracons hrough shared srucural characerscs. 3

4 4 Km and Peea nformaon flo neork G = (N,C, A, M ) ner-vehcle communcaon even Dnamcs of nformaon flo ner-vehcle communcaon neork C G = (C,M) VD Traveled lnk Lnk enrance me Travel me :55:33 55seconds Varable name Vehcle denfcaon (VD) number Traveled lnk Lnk enrance me Lnk ravel me Eample of ravel eperence daa Dealed represenaon D of he vehcle A lnk ha he vehcle passes hrough Tme ha he vehcle eners he lnk Vehcle ravel me beeen he o nodes of a lnk Node n he raffc flo neork Travel eperence daa generaon even A par of ner-vehcle communcaon nodes Travel eperence daa (TED) node Correspondng node generaon n he nformaon flo neork Locaon of vehcles Traffc flo neork T G = (N,A) Travel eperence daa generaon even Dnamc vehcle rajecor ner-vehcle communcaon even ner-vehcle communcaon lnk A par of vrual nervehcle communcaon (VC) nodes Lnks n nformaon flo neork Fg. 1. Concepual frameork of he negraed mul-laer neork frameork Noaon The follong noaon s used o represen varables n he negraed mul-laer neork frameork. T Traffc flo neork G = (N,A) N : he se of phscal nodes A : he se of phscal lnks X : he se of vehcles : subscrp for a vehcle, X T : he duraon of neres n he V2V-based ATS : superscrp for (connuous) he me of neres, [0,T] : a phscal node n he neork, N (, j ) : a phscal lnk n he neork, (, j) A C ner-vehcle communcaon neork G = (C,M) C : he se of ner-vehcle communcaon nodes M : he se of ner-vehcle communcaon lnks : broadcasng ner-vehcle communcaon nodes for vehcle, C : recevng ner-vehcle communcaon nodes for vehcle, C (, ) : ner-vehcle communcaon lnk from vehcle o vehcle, (, ) M nformaon flo neork G = (N,C, A, M ) N C P A M : he se of ravel eperence daa (TED) nodes : he se of vrual ner-vehcle communcaon (VC) nodes : he se of nodes n he nformaon flo neork, P {N,C }, N C : he se of nformaon flo propagaon rajecor lnks (T-lnk) ndcang he vehcle rajecor drecon based on he raffc flo : he se of ner-vehcle communcaon based nformaon flo propagaon lnks (-lnk) denong he drecon of nformaon flo based on he ner-vehcle communcaon : ravel eperence daa (TED) node ndcang a ravel eperence daa generaed b vehcle a node a me, X, N, N : vrual ner-vehcle communcaon (VC) node denong ha vehcle broadcass ravel eperence daa a me, C : vrual ner-vehcle communcaon (VC) node

5 Graph-based Modelng of nformaon Flo Evoluon and Propagaon under V2V Communcaons based Advanced Traveler nformaon Ssems 5 denong ha vehcle receves ravel eperence daa a me, C p, q : nodes n he nformaon flo neork assocaed h vehcle a me ; p, q P ( p 1,q 2 ): nformaon flo propagaon rajecor lnk assocaed h he rajecor drecon of vehcle from me 1 o me 2, 1 2 p, q P, X, (, ) : ner-vehcle communcaon based nformaon flo propagaon lnk represenng he drecon of nformaon flo (from vehcle o vehcle ), 1 2 (, ) C,, Phscal raffc neork Under V2V-based ATS, he equpped vehcles generae daa on her ravel eperences usng a global posonng ssem (GPS) and a dgal neork mappng. Thereb, hen a vehcle reaches he end of he lnk (ha s, he assocaed donsream node n he raffc neork), generaes ravel eperence daa. As shon n Fg. 1, he ravel eperence daa of a vehcle ncludes s vehcle denfcaon (VD) number, he denfcaon number of he lnk raversed, he lnk enrance me, and he lnk ravel me. Snce he raffc flo neork s a phscal en, e assume ha he generaon of ravel eperence daa based on vehcle rajecores are observable, and gven n he sud. Le G T = (N,A) denoe a raffc flo neork n hch vehcles have an abl o communcae h each oher. A se N of nodes corresponds o phscal nersecons or desgnaed pons n he raffc flo neork, and a se A of dreced lnks corresponds o road lnks. Ths laer capures he spaoemporal neracons among vehcles hrough her rajecores. The me-dependen locaons of vehcles deermne he evens of neres; he ravel eperence daa generaon n he raffc flo neork, and he feasbl of ner-vehcle communcaon based on relevan echncal consrans n he ner-vehcle communcaon neork. The ravel eperence daa generaed n he phscal raffc neork represen one of he componens used o consruc he nformaon flo neork. 2.3 Vrual ner-vehcle communcaon neork The knon me-dependen locaons of vehcles n he raffc flo neork and he ner-vehcle communcaon consrans are used o consruc he ner-vehcle communcaon neork. An ner-vehcle communcaon even s veed as a one-a ransfer from a broadcasng vehcle o a recevng vehcle. The ransfer of nformaon hrough ner-vehcle communcaon creaes a vrual communcaon node se C and a dreced communcaon lnk se M, leadng o he C ner-vehcle communcaon neork G = (C, M). The vrual ner-vehcle communcaon neork represens he characerscs of he ner-vehcle communcaon evens (ha s, hch vehcle broadcass and hch vehcle receves he nformaon n a gven even), as llusraed n Fg. 1. Based on he occurrence of evens n he nervehcle communcaon neork, nformaon flo propagaon akes place beeen vehcles. The communcaon range llusraes he phscal dsance hn hch V2V communcaon can poenall occur. The communcaon sgnal poer decreases h dsance. Hence, s assumed ha V2V communcaon ll no occur ousde he specfed range. To equpped vehcles can poenall communcae h each oher hen phscal dsance s less han a predefned communcaon range r. Hoever, mulple rasmssons from vehcles hn communcaon range leads o nerference, hch ma resul n he falure of recevng nformaon from oher vehcle. The nerference rae s defned as follos (Gupa and Kumar, 2000): T T (1) / ( E ) 2 2 X here, s defned as he GPS locaon coordnae of an equpped vehcle hn communcaon range, X. We assume ha he poer levels of vehcles ( T and T ) are dencal and he amben nose poer level ( E ) s ero. Sgnal poer decas h dsance and he vehcle ll succeessfull receve he nformaon from vehcle f sasfes he mnmum sgnal-o-nerference rao of (he sud epermens use = 2 based on Gupa and Kumar, (2000)). Specfcall, all equpped vehcles posons hn communcaon range of vehcle n he raffc neork are racked. The me nerval of V2V communcaon s se o 0.5 seconds and he accomplshmen of ner-vehcle communcaon beeen and hose vehcles s checked ever nerval. Consder he locaons of vehcles shon n Fg. 2, here vehcle s broadcasng and s he recevng vehcle. Vehcles 3, 4 and 5, hose posons are hn communcaon range r from vehcle, can poenall nerfere h he communcaon from vehcle from vehcle. Ths nerference rae s calculaed based on Equaon (1). The banddh (capac) of ner-vehcle communcaon s a lmng facor and can resul n dropped daa packes. n hs sud, e assume a 2Mbps daa ransmsson rae and 0.5 seconds for frequenc of communcaon. Ths s appled b resrcng he number of ravel eperence daa o be broadcas n each ner-vehcle communcaon. The ner-vehcle communcaon laer uses he rajecores of all equpped vehcles from he raffc flo neork, and compues heher vehcles succeed or fal o

6 6 Km and Peea communcae h each oher n he presence of he nervehcle communcaon consrans dscussed hereofore. Hence, he rerospecve capabl n he proposed frameork s deermnsc n he sense ha gven he dnamc vehcle rajecores from he raffc flo neork and he ner-vehcle communcaon consrans, nervehcle communcaon evens are deermnscall compued. Gven ha hs deermnsc frameork s a buldng block, ongong ork b he auhors seeks o provde a sochasc capabl for he ner-vehcle communcaon laer o model he effecs of he nervehcle communcaon consrans n erms of capurng he randomness relaed o he V2V communcaon. nformaon flo evoluon and propagaon, and he assocaed evoluon of vehcle knoledge. VD Traveled Lnk Travel me lnk enrance me :55:33 55seconds Travel eperence daa sored n vehcle :45: :02:18 63 Travel eperence daa ransmed from vehcle :03:21 41 Travel eperence daa generaed b vehcle 33 GPS VD Traveled lnk Lnk enrance me Travel me :02: :01: :55: :45: :02: :03:21 41 Travel eperence daa ransmed from vehcle 33 (a) Eample of vehcle knoledge (vehcles 33 and 25) and he assocaed ravel eperence daa X : vehcle 2 2 : GPS locaon coordnae of an equpped vehcle VD 17 ner-vehcle communcaon Temporar memor Sore VD 25 : ner-vehcle communcaon : nerference : vehcle s vald communcaon range r Fg. 2. nerference among vehcles. 2.4 Graph-based represenaon of nformaon flo neork Under a V2V based ATS, a vehcle connuousl updaes s knoledge usng s on eperence and he anonmousl obaned ravel eperence daa of oher vehcles. These ravel eperence daa are sored n he emporar memor on board he vehcle s ssem, and duplcae (spaoemporal daa of he same vehcle) and/or older (daa older han 30 mnues n he sud epermens) daa are dscarded. Fg. 3 llusraes he vehcle knoledge evoluon due o ravel eperence daa generaon and ner-vehcle communcaon, and deals of he assocaed daa packe confguraon. Fg. 3(a) shos he deals of he vehcle knoledge of vehcles 33 and 25, and he assocaed ravel eperence daa. Fg. 3(b) llusraes ho evens (generaon of ravel eperence daa, and ner-vehcle communcaon) mpac he evoluon of vehcle knoledge n Fg. 3(a). For eample, he vehcle knoledge of vehcle 33 consss of ses of daa ha generaes and sores, and receves from oher vehcles (vehcle 17 n Fg. 3). The nformaon flo neork, hose flos are a se of ravel eperence daa, s consruced usng nodes and lnks o map ha/hen/here nformaon s generaed and ho propagaes. llusraes he dnamc naure of he nformaon flo from one vehcle o anoher VD 33 nformaon flo hn he vehcle Vehcle ravel eperence daa (b) llusraon of he ravel eperence daa generaon and ner-vehcle communcaon Fg. 3. Vehcle knoledge evoluon. The nformaon flo neork G = (N, C, A, M ) has o pes of nodes: a ravel eperence daa (TED) node N generaed b he correspondng each even of ravel T eperence daa generaon n G, and ) a par of vrual ner-vehcle communcaon (VC) nodes (one for broadcas and he oher for recevng) C represenng he correspondng each even of ner-vehcle communcaon n C G, as llusraed n Fg. 1. To ses of lnks represen he dnamcs of nformaon flo evoluon and propagaon. The dreced nformaon flo propagaon rajecor lnks (T-lnk) A denoes he spaoemporal rajecores of he same vehcles hrough TED-TED, TED-VC, VC-TED or VC-VC node connecons. The ner-vehcle communcaon based nformaon flo propagaon lnks (-lnk) M connec each par of nodes (VC-VC) correspondng o nervehcle communcaon evens beeem o vehcles nformaon flo generaaon/deleon When a V2V-equpped vehcle reaches a phscal nersecon, generaes daa on he ravel me eperenced on he lnk jus raversed. Ths even s denoed b a TED node, denoed as N. represens he ravel eperence daa generaed b vehcle X a node a me. As llusraed n Fg. 3, he ravel eperence daa consss of he vehcle denfcaon number, he lnk enrance me, and he eperenced lnk ravel me. Fg. 4 shos he roue

7 Graph-based Modelng of nformaon Flo Evoluon and Propagaon under V2V Communcaons based Advanced Traveler nformaon Ssems 7 rajecor of vehcle X, and he correspondng TED 1 nodes 2 and j generaon n he nformaon flo neork. The share he same opolog of he phscal nodes N and j N, bu a me pons and 1, 2 respecvel. Spaoemporal vehcle rajecor of vehcle Vehcle locaed a node j a me 2 j 2 j 2 k 3 Node n he raffc flo neork Vehcle Trajecor of vehcle n he raffc flo neork k 3 Travel eperence daa generaed b vehcle from node j a me 2 VD Traveled Enrance Lnk ravel lnk Tme me (, j) l 4 l G T G Travel eperence daa (TED) node Correspondng TED node generaon n G Fg. 4. Travel eperence daa nodes n he nformaon flo neork correspondng o evens of ravel eperence daa generaon n he phscal raffc neork. Snce he TED nodes are generaed based on vehcles me-dependen locaons n he raffc flo neork, he TED node can characere he spaoemporal dnamcs of 1 raffc flo. Thus, also characeres he spaoemporal vehcle rajecor for vehcle X locaed a node a me. Therefore, n Fg. 4, he TED node j 2 n he nformaon flo neork denoes he ravel eperence daa ha s generaed b a he phscal node j N a me 2. As dscussed earler, f he daa sored n a vehcle s onboard memor sorage becomes large, can burden he ner-vehcle communcaon of daa and lead o falure n erms of communcang ha daa. To avod hs ssue, older daa s removed from he vehcle s onboard memor sorage b desgnang as he mamum daa sorage me nerval (for eample, 30 mnues n he sud epermens) nformaon flo evoluon and propagaon We consruc he VC nodes and he dreced lnks (Tlnks and -lnks) n he nformaon flo neork o represen ho nformaon flo evolves and propagaes. When vehcle broadcass s ravel eperence daa o 1 2 vehcle, a par of VC nodes and C s generaed n G correspondng o he ner-vehcle communcaon C nodes ( and ) of G. The VC node pars conss of he broadcasng VC node C and he recevng VC node b C r ( C, b Cr C ). Each par of VC nodes s conneced b an ner-vehcle communcaon based nformaon flo propagaon lnk (lnk), (, ) M n G, hch s defned as 1 2 follos: 1 2 M {(, ) C b C r, 1 2} (2) A dreced -lnk connecs a par of VC nodes from he broadcasng VC node o he recevng VC node C r C b. Ths represenaon sores he correspondng nformaon flo propagaon hrough he ner-vehcle communcaon. We asume ha he broadcasng and recevng of nformaon sar a he same me. j 2 j j k 5 Node n raffc flo neork Trajecor of vehcle n he raffc flo neork Travel eperence daa (TED) node A par of vrual nervehcle communcaon (VC) nodes Correspondng TED node generaon n G Correspondng VC node generaon n G k l 8 l 6 ner-vehcle communcaon lnk G C G T G ner-vehcle communcaon consrans A par of ner-vehcle communcaon nodes nformaon flo propagaon rajecor (T)-lnk ner-vehcle communcaonbased nformaon flo propagaon ()-lnk Fg. 5. Represenaon of nformaon flo evoluon and propagaon. Fg. 5 llusraes he generaon of he VC nodes and - lnks n he nformaon flo neork correspondng o he ner-vehcle communcaon. For eample, he occurrence of he ner-vehcle communcaon from vehcle o vehcle a me 7 s represened as he par of 7 broadcasng VC node 7 and recevng VC node, 7 7 and he dreced -lnk (, ). These VC nodes and -

8 8 Km and Peea lnks can map hen/here nformaon propagaes from one vehcle o anoher vehcle. We defne a se of nformaon flo propagaon rajecor lnks (T-lnks) A as follos: 1 2 A {( p, ) (P P q ) X, 1 2} (3) As he TED and VC nodes for a vehcle occur along h s rajecor, a dreced T-lnk connecs TED-TED, TED- VC, VC-TED or VC-VC nodes based on he rajecor of vehcle from me o me 1 2 ( 1 2 ). Hence, a se of T-lnks represens he assocaed spaoemporal raffc flo dnamcs. Fg. 5 llusraes ha he dreced T lnks (, j ), ( j, ) and (, ) connec he TED- TED, TED-VC, and VC-VC nodes based on he vehcle rajecor drecon. These T-lnks eplan ho nformaon flo propagaes along h hs vehcle s rajecor. 2.5 Graph srucure and reverse search algorhm for vehcle knoledge denfcaon The me-dependen knoledge of a vehcle a a parcular locaon (ndcaed b a specfc TED node) s represened b a conneced group of TED nodes (nerpreed as ravel eperence daa) and he assocaed dreced lnks n he graph srucure of he nformaon flo neork. Thereb, deermnng a subgraph ha s conneced o a specfc node n he nformaon flo neork denfes he vehcle knoledge of neres. We formulae hs process as a searchng problem. A graph-based reverse search algorhm (Ahuja e al., 1993) racks he flo of nformaon usng a backrackng logc from he specfc node, b racng he drecon oppose o ha of he dreced lnk and denfng each source of nformaon (ha s, each TED node). The me-dependen knoledge of a vehcle locaed a a TED node can be denfed b denfng all nodes ha can reach along dreced pahs. Fg. 6 llusraes he procedure o mplemen he reverse search algorhm o denf he vehcle knoledge. n he nalaon sep, ever node s se as unmarked, and Travel Daa, hch collecs all nodes ha are reachable from a specfc node s (vehcle locaon), s se o emp. Here, a node l s reachable from anoher node k f here s a dreced pah from k o l. A specfc node j = s s marked nall and added o he emp ls Ls. We fan ou from j o denf nodes ha can reach. To do so, e search for admssble lnks ha are ncden o j. A lnk (, j) s referred o as admssble f node s unmarked and node j s marked. For each admssble lnk, e desgnae s unmarked node as vsed and ag as marked. Node s added o Ls and Travel Daa. Afer all admssble lnks for j are scanned, remove j from Ls. Go o he ne node n Ls and repea he aforemenoned algorhmc process unl here are no nodes n Ls. When he algorhm ermnaes, each TED node n Travel Daa ndcaes a generaed ravel eperence daa and each VC node eplans ho s obaned hrough ner-vehcle communcaon. begn end; Unmark all nodes n G ; Mark node s ; Ls:={s}; hle Ls 0 do begn selec a node j n Ls; hle here s an admssble lnk (, j) ha s ncden o node j do begn mark node ; add node o Ls; add node o Travel Daa; end; delee node j from Ls; end; Fg. 6. mplemenaon of he graph-based reverse search algorhm. 3 RETROSPECTVE MODELNG CAPABLTY N THE GRAPH-BASED FRAMEWORK 3.1 Vehcle knoledge updae n smulaon-based approach Modelng a large-scale V2V-based ATS s nherenl comple. Hence, a smulaon-based approaches has pcall been used o denf vehcle knoledge n V2Vbased ATS sudes. Pas sudes (Wu e al., 2005; Km e al., 2009; Km, 2010) use raffc smulaors as he raffc flo laer, and ner-vehcle communcaon consrans govern daa echange based on each equpped vehcle s locaon nformaon n each smulaon me sep. Thereb, n each me sep of he smulaon-based approach, each vehcle s knoledge s updaed usng raffc daa receved from oher vehcles as ell as he daa generaed b he vehcle self. n an updae process, each vehcle s knoledge afer a relevan even s coped o an ndvdual memor locaon (for ha vehcle) hch sores s prevous vehcle knoledge; hs s llusraed n Fg. 7 for o me seps ( 1 and 2). Whle hs s an nuve approach ha mmcs he ndvdual vehcles onboard memor sorage and copes he ravel eperence daa from one vehcle o anoher, lacks rerospecve modelng capables o arculae eplcl ho nformaon flo evolves and propagaes. Furher, as eplaned n Secon 3.3, an eensve updae process s requred o updae each vehcle s knoledge under hs approach.

9 X1 W1 X2 W2 X3 X4 X j Graph-based Modelng of nformaon Flo Evoluon and Propagaon under V2V Communcaons based Advanced Traveler nformaon Ssems X1 W1 Y1 X2 W2 Y2 X3 X4 j Daa sorage of vehcles a me 1 X Y W Z Z1 Y1 Y2 Y3 Y1 Y2 Y3 W k k Z1 W1 X1 W2 X2 W3 Daa sorage of vehcles a me X Y W Z X Y W1 X1 W2 X2 W3 l l Z1 Generaon of ravel eperence daa across sroage locaons Z1 Z2 Z3 Copng of he ravel eperence daa across sorage locaons Z Y Z2 Z3 2 Travel eperence daa generaon even Correspondng ravel eperence daa generaon n sorage locaon ner-vehcle communcaon even Correspondng copng of he ravel eperence daa across sorage locaons Fg. 7. Vehcle knoledge updae n he smulaon-based approach. 3.2 Rerospecve modelng capables Rerospecve modelng of he negraed mullaer neork There s a need o undersand and model he dnamcs of nformaon flo eplcl a he mul-laer neork level. Hoever, he negraon of he nformaon flo evoluon and propagaon h he raffc flo dnamcs and nervehcle communcaon nroduces sgnfcan comple for o reasons. Frs, he need o rack vehcles from he perspecve of nformaon flo evoluon and propagaon n addon o he phscs of vehcular neracons requres a mul-laer neork approach. Second, snce nformaon s echanged b vehcles connuousl, here s a need o ensure conssenc n nformaon flo propagaon over space and me. n hs cone, he evoluon of vehcle knoledge can be analed n he mul-laer neork frameork hrough he nformaon flo neork b lnkng he TED nodes o evens n he raffc flo neork and VC nodes o he evens n he ner-vehcle communcaon neork, hereb llusrang he dnamcs of raffc flo and he occurrence of he ner-vehcle communcaon, respecvel. Fg. 8(a) shos an eample o rack he evoluon of knoledge of o vehcles h smlar roues based on he dnamcs of raffc flo and he ner-vehcle communcaon evens. B undersandng ha ner-vehcle communcaon occurs h hom, hen vehcles communcae, and ha vehcle knoledge s ransmed over space and me, nsghs can be generaed on he neracon of evens ha nfluence he evoluon of vehcle knoledge. Fg. 8(b) llusraes a spaoemporal analss of he knoledge of o vehcles shon n Fg. 8(a) n space and me. llusraes ha dnamc vehcle knoledge enals he follong aspecs: dnamc spaoemporal coverage, me dela (dfferen color shades mpl dfferen me delas), qual and quan of ravel eperence daa (hckness of lne ndcaes he number of ravel eperence daa for a correspondng lnk), and he relevance of daa for s rp (based on curren locaon and desnaon). These observaons llusrae he need for a ssemac undersandng of vehcle knoledge characerscs o leverage s use o develop drver roue gudance sraeges, and ssem operaor sraeges for he effcen spread of useful nformaon nformaon propagaon chan The graph srucure of he nformaon flo neork provdes rerospecve nformaon relaed o ho nformaon evolves and vehcle knoledge s updaed. Ths provdes a capabl o rack he spaoemporal characerscs of nformaon flo evoluon and propagaon drecl hrough a conneced graph srucure and generae a fundamenal undersandng of ho evens affec he evoluon of he nformaon flo neork. Fg. 9(a) llusraes an eample of an nformaon flo neork usng he graph-based represenaon and a subgraph of G ndcang vehcle knoledge. The vehcle knoledge of hs vehcle consss of a se of subgraphs ha generaes or receves from oher vehcles (from vehcle and ) hrough ner-vehcle communcaon. Therefore, he evoluon of vehcle knoledge of neres can be racked from an pon usng a graph-based search algorhm. A subgraph conneced o a VC node represens vehcle knoledge receved from anoher vehcle hrough each ner-vehcle communcaon. As shon n Fg. 9(a), he vehcle knoledge of vehcle (represened b a subgraph usng doed lnes) s ransmed o vehcle hrough he ner-vehcle communcaon from vehcle o vehcle. Fg. 9(b) llusraes he Travel Daa obaned usng he graph-based search algorhm. The algorhm fans ou from he node n he shaded crcle n Fg. 9(a) n sequence, denfng he reachable nodes n Travel Daa. VC nodes n Travel Daa form an nformaon propagaon chan hch can address he follong: (1) hch vehcle conrbues o propagang he ravel eperence daa of vehcle o vehcle?, and (2) ha s he me requred o propagae nformaon o he arge vehcle hrough nervehcle communcaon? Fg. 9(c) shos each vehcle s knoledge a he end of he smulaon based on he same evens as n Fg. 9(a). As can be seen, unlke n Fg. 9(b), he smulaon-based approach s lmed n s abl o llusrae he nformaon flo evoluon and propagaon. Tha s, Fg. 9(c) shos he vehcle knoledge hou ndcang he me dmenson and he nformaon propagaon chan. canno easl nfer hch vehcle 9

10 10 Km and Peea conrbues o propagang he ravel eperence daa n he shaded square and hen hs eperence daa s obaned b vehcle. Ths s because he smulaon-based approach copes daa from he vehcle knoledge of one vehcle o anoher hou eplcl ndcang hch vehcle s nvolved n he ner-vehcle communcaon. Therefore, rackng he spaoemporal characerscs of he nervehcle communcaon requres he aggng of hen such communcaons occur and hch vehcles are nvolved n such communcaon, hch can be compuaonall epensve. Fg. 10(a) llusraes a subgraph of G ndcang he propagaon of a sngle un of nformaon. A subgraph can eplcl address hen and o hom a specfc un of nformaon propagaes. For eample, ravel eperence daa generaed b vehcle a me 0 propagaes o vehcle and hrough he ner-vehcle communcaon, and he even locaons can be racked usng he TED node as dscussed n Secon Dnamc vehcle knoledge ner-vehcle communcaon 3 = =16 1 =11 Dnamc vehcle rajecor 3 = =12 1 =8.5 Vehcle A a me =16 Vehcle B a me =12 Vehcle A Vehcle B : radus of crcle corresponds o he number of ravel eperence daa a each me pon : vehcle rajecor (avalable hrough TED nodes) * : locaon of ner-vehcle communcaon even (avalable hrough VC nodes) 1 =8.5: ndcaes he me vehcle arrves a ceran pon (a) Rerospecve capabl o address he neracons h oher o laers Tme dela Daa sampe se mn Locaon of vehcle (b) Dnamc vehcle knoledge of o vehcles h smlar roues Fg. 8. Mul-laer neork analss and evoluon of vehcle knoledge Vehcle locaed a node a me a me a me Propagaon chan a me 2 (b) llusraon of Travel Daa n he reverse search algorhm ner-vehcle communcaon beeen vehcle and (a) llusraon of graph srucure and s reverse search algorhm Vehcle knoledge of vehcle locaed a node a me A subgraph of G Knoledge of all vehcles sored under he daa srucure of smulaon-based approach Copng of he ravel eperence daa across sorage locaons Daa sorage of vehcle (c) Smulaon-based approach Fg. 9. Comparson of spaoemporal rackng capables. 1 TED node generaed b vehcle VC node generaed b vehcle T-lnk -lnk Reverse search Vehcle of neres/ nformaon of neres The order n hch search algorhm vss nodes Travel eperence daa generaed b vehcle

11 Graph-based Modelng of nformaon Flo Evoluon and Propagaon under V2V Communcaons based Advanced Traveler nformaon Ssems Travel eperence daa generaed b vehcle a node j a me 0 Vehcle of neres/ Forard search nformaon of neres (a) llusraon of graph srucure and s forard search algorhm Vehcle locaon ha has receved a specfc nformaon : Locaon here he nformaon s nall generaed (b) Dnamcs of nformaon flo propagaon (under marke peneraon of 10% and 20%, respecvel) A subgraph of G Afer 25mnues Afer 15mnues Afer 5mnues Fg. 10. Dnamc nformaon flo evoluon and propagaon under dfferen marke peneraon raes. Therefore, he graph srucure ransparenl provdes a drec lnk o rack he nformaon flo evoluon and propagaon from one vehcle o anoher usng he graphbased search algorhm from a TED node (hch ndcaes a specfc ravel eperence daa). B conras, he smulaonbased approach requres a scan of he knoledge of all vehcles sored usng he daa srucure n Fg. 7 o deermne heher has a parcular ravel eperence daa of neres, hch can be compuaonall epensve. Fg. 10(b) shos he propagaon of a ravel eperence daa from a vehcle n erms of he dnamcs of raffc flo and ner-vehcle communcaon evens. shos hch vehcle receves hs nformaon over me (5, 15, and 25 mnues afer s frs generaed) under dfferen marke peneraon raes (10% and 20%). 3.3 Effcenc of he graph daabase Graph daabase n he graph-based approach n conras o he daa sorage mechansm of he smulaon-based approach llusraed n Secon 3.1, he graph daabase used n hs sud shares daa n a sngle memor o represen he knoledge of all vehcles hrough nerconneced nodes and lnks, and uses a local search o denf he vehcle knoledge of an vehcle. A reverse graph-based search algorhm leverages hs srucure o raverse he graph n a drecon oppose o ha of he nformaon flo propagaon from he curren locaon of a vehcle o deermne he curren vehcle knoledge. More deals on graph daabases can be found n Robnson e al. (2013) and Sakr and Pardede (2011) Memor usage effcenc B sharng he nodes and lnks o represen he knoledge of vehcles n he nformaon flo neork, he graph daabase leads o memor se effcenc compared o a smulaon-based approach hch needs o sore he vehcle knoledge separael for each vehcle o updae he knoledge of oher vehcles. The reverse search algorhm for each vehcle of neres denfes he subgraph of he nformaon flo neork ha denoes s knoledge. Thereb, he proposed graph daabase s effcen and elmnaes he need o sore vehcle knoledge for each vehcle. B conras, he smulaon-based approach adaps an updae mechansm hch copes daa from he mos recen vehcle knoledge of he broadcasng vehcle o he recevng vehcle a he me of ner-vehcle communcaon, as seen n Fg. 7. Thus, he smulaonbased approach mus memore all vehcles knoledge ndvduall o compue oher vehcles knoledge. Ths leads o an eponenal ncrease n memor se usage for large scale real-orld raffc neorks. Le X be he number of equpped vehcles n V2Vbased ATS, and N, C, A, and M, he number of elemens n N, C, A, and M, respecvel. Snce he graph daabase of he nformaon flo neork s sored as a sngle graph, here s no duplcaon of nodes. Hence, he memor usage for represenng he nformaon flo neork can be bounded b ( N + C + A + M ). Hoever, he memor usage for he smulaon-based approach requres up o ( N X ). Ths s because sores all vehcles knoledge ndvduall o compue oher vehcles knoledge. Tha s, each of he X vehcles can have a mamum of N ravel eperence daa nodes Compuaonal effcenc n he graph daabase, he vehcle knoledge of an vehcle can be denfed b vsng onl conneced subgraph of nformaon flo neork. Therefore, does no requre an updae of he knoledge of all vehcles o denf a specfc vehcle s knoledge. n addon o addressng he hree real-orld objecves denfed n Secon 1, hs provdes a sgnfcan praccal benef n erms of he abl o rack he spaoemporal evoluon of nformaon for a specfc class of vehcles. Furher, raversng from one node o anoher s a consan me operaon. Thus, he search me s defned solel b he number of nodes and lnks denfed b he local seach. 11

12 12 Km and Peea Ths s rrespecve of he se/opolog of he nformaon flo neork as a hole. The me akes o search s deermned b he local opolog of he subgraph surroundng he parcular node beng raversed from. Thus, he graph-based approach can denf a vehcle s knoledge h comple O ( A + M ). n he reverse search algorhm n Fg. 6, he admssble lnks are denfed and ne nodes are marked and added o Ls and Travel Daa. The effor spen n denfng he admssble lnks from each node j n he nformaon flo neork s equal o he number of adjacen lnks o j, ndcaed b he lnk adjacenc ls of j. Hence, he graphbased approach for a vehcle of neres a a specfc locaon has he order of he oal number of lnks n G ;O ( A + M ). Then, for a specfc class of vehcles (such as vehcles n a ceran geographcal area or vehcles eposed o a ceran varable message sgn), he algorhm has an order O (( A + M ) K ), here K s he number of vehcles of he specfc class K. B conras, o rack he knoledge of a specfc class of vehcles, he smulaon-based approach reles on updang all processes sarng from he frs me nerval o updae he knoledge of a vehcle. Thus, he smulaon-based approach n Fg. 7 can denf a vehcle s knoledge of neres onl hen all processes of copng he ravel eperence daa across sorage locaons are compleed. Therefore, he eecuon me o denf he knoledge of a sngle vehcle s he same as ha for denfng he knoledge of all vehcles. The compuaonal comple s deermned as follos. For each vehcle, each ravel eperence daa s added o s sorage locaon. Snce here are N such ravel eperence daa generaed, he compuaonal me o sore all such daa has an order O ( N ). When ner-vehcle communcaon akes place, he vehcle knoledge s coped from he broadcasng vehcle sorage locaon o he recevng vehcle sorage locaon. Snce he mamum se of vehcle knoledge has an order O ( N ), he cop operaon compuaonal me for each ner-vehcle communcaon has an order O ( N ). As he number of ner-vehcle communcaons s N, he me comple of updang all vehcles knoledge n he smulaon-based approach s O ( N + N M ). Snce M can be reall large, hs approach s less effcen compared o he graph-based approach. Furher, he ravel eperence daa are sored n he emporar memor on board he vehcle s ssem, and duplcae daa from man oher vehcles are processed and dscarded. n hs cone, he smulaon-based approach copes he ravel eperence daa across sorage locaons and eecues he process o fler duplcae daa. Hoever, n he graph daabase, each ravel eperence daa s represened b a node onl once and s propagaon s capured hrough he lnk represenaon, elmnang hese flerng processes. Ths observaon s mporan o furher llusrae he effcenc of he graph-based approach. 4 NUMERCAL EXPERMENTS 4.1 Epermen desgn The Borman Epressa neork n norhern ndana, hch ncludes nersaes 80/94 and 65, and he surroundng arerals, s used as he sud neork. consss of 197 nodes and 460 lnks, and has a se 11.3mles 8.5mles. Dfferen demand levels (lo: 11,074 vehcles; medum: 43,988 vehcles; hgh: 88,123 vehcles) and marke peneraon raes (1%, 5%, 10% and 20%) ha denoe percenage of V2V-equpped vehcles, are consdered for analss over a 90-mnue perod of neres. Fg. 11 shos he Borman Epressa neork. Fg. 11. Sud neork Of he hree laers of he V2V-based ATS, he raffc flo laer s replcaed usng a mesoscopc vehcular raffc smulaor, DYNASMART-P, o represen he flo dnamcs n he raffc neork. To ensure conssenc n comparson, boh he graph-based and smulaon-based approaches use he correspondng rajecores of he V2Vequpped vehcles n he raffc flo laer for a 90-mnue me horon of neres, and he ner-vehcle communcaon consrans, o deermne he vehcle knoledge. The follong ner-vehcle communcaon consrans are used: he ner-vehcle communcaon range (250m), nerference rae (Equaon (1)), and banddh (2Mbps). Snce he graph-based represenaon of he nformaon flo neork does no sore he ravel eperence daa n dsparae sorage locaons, o calculae he daa packe se of a vehcle he banddh consran s appled b resrcng he number of ravel eperence daa (12,500 TED nodes) o be searched from a specfc node n he nformaon flo neork. Correspondngl, for conssenc, he smulaon-based approach s alloed o cop onl a mamum number of 12,500 such ravel eperence daa for each ner-vehcle communcaon. The

13 Graph-based Modelng of nformaon Flo Evoluon and Propagaon under V2V Communcaons based Advanced Traveler nformaon Ssems performance of he o approaches o denf he vehcle knoledge under varous scenaros s eamned n erms of memor usage effcenc and compuaonal effcenc. n all scenaros, demand s loaded durng he frs 90 mnues of analss. We eamne elve scenaros ha represen combnaons of hree demand levels (lo, medum, and hgh) and four marke peneraon raes (1%, 5%, 10%, and 20%). 4.2 nformaon flo neork se and consrucon me Gven evens n he raffc flo and ner-vehcle communcaon neorks, he nformaon flo neork s consruced n he graph-based approach b addng ne nodes and lnks. The ses of he nformaon flo neork under he 250m communcaon range, and dfferen scenaros of demand and marke peneraon levels are summared n Table 1 n erms of he number of equpped vehcles X and he number of neork componens, ncludng TED nodes N, VC nodes C, T-lnks A, and -lnks M. Here, N ncreases h he number of equpped vehcles X. As vehcles are more lkel o echange nformaon under hgher demand and marke peneraon raes, he assocaed number of nodes C, and lnks A and M ncrease rapdl. The me o consruc G n erms of generang he node-lnk adjacenc mar of he nformaon flo neork s also shon n Table 1. Gven he nformaon flo neork for each scenaro, he graph-based approach denfes he spaoemporal characerscs of vehcle knoledge usng he graph-based reverse search algorhm. 4.3 Performance evaluaon Memor usage effcenc Table 2 compares he memor usage of he graph-based nformaon flo neork and he smulaon-based approach. For eample, for he hgh demand scenaro h marke peneraon 20%, he memor usage o sore he knoledge of all vehcles requres 2.9 MB for he graphbased represenaon and 1.45 GB for he smulaon-based approach. The assocaed raos of he memor usage of he smulaon-based and graph-based approaches are shon n he las column. As he number of ner-vehcle communcaon evens ncreases rapdl h hgher demand and marke peneraon raes, he memor requred o sore he knoledge of all vehcles ndvduall ends o ncrease eponenall. As dscussed earler, he superor memor usage effcenc under he graph-based approach s due o he sharng of nodes and lnks o represen vehcle knoledge raher han duplcang he ravel eperence daa hen nformaon propagaon occurs. Table 1 nformaon flo neork se and consrucon me under dfferen scenaros Scenaro* Se of he nformaon flo neork Neork consrucon me X N C A M (un: seconds) L , , L ,060 4,146 8,663 2, L-10 1,091 10,324 15,936 25,169 7, L-20 2,173 20,763 62,280 80,870 31, M ,490 2,488 5,606 1, M-5 1,834 16,885 54,464 69,515 27, M-10 3,682 33, , , , M-20 7,354 67, , , , H ,959 8,260 12,618 4, H-5 2,986 24, , ,994 87, H-10 5,953 49, , , , H-20 11,953 99,947 1,945,198 2,033, , *L: lo demand, M: medum demand, H: hgh demand; 1,5,10, and 20: marke peneraon rae (n percen) Compuaonal effcenc Table 3 compares he compuaonal mes of he graphbased and he smulaon-based approaches o deermne he vehcle knoledge of all vehcles a her desnaons o ensure conssenc n comparson, as he smulaon-based approach can deermne her vehcle knoledge onl a he end of he horon of neres (90 mnues n he sud epermens). The number of equpped vehcles X, shon n he second column, represens he number of mes ha he reverse search algorhm s eecued. Snce he graphbased approach can denf he knoledge of a vehcle of

14 14 Km and Peea neres hou updang he oher vehcles knoledge, he assocaed average compuaonal mes across he vehcles n column 2 are shon n he hrd column. The compuaonal mes o deermne he knoledge of all vehcles are shon n he fourh column (as he produc of he second and hrd columns). The compuaonal mes o deermne he knoledge of all vehcles under he smulaon-based approach are shon n las column. n summar, an eplc modelng of he nformaon flo neork as a graph daabase leads o memor effcenc due o he crcumvenon of redundan sorage of he ravel eperence daa. Furher, he local searchng of onl reachable nodes, hch are a subgraph of he nformaon flo neork, enables he compuaonal effcenc. Table 2 Memor usage o updae vehcle knoledge Scenaro* Memor usage (un: Megabes) Graph-based approach Smulaon-based approach Comparson rao L L L L M M M M H H H H , *L: lo demand, M: medum demand, H: hgh demand; 1,5,10, and 20: marke peneraon rae (n percen) Scenaro* Table 3 Compuaon me o deermne vehcle knoledge for all vehcles Graph-based approach for he vehcle knoledge denfcaon Smulaon-based approach Number of equpped vehcles X Average eecuon me of he search algorhm (seconds) Tme o deermne he knoledge of all vehcles (seconds) Tme o deermne he knoledge of all vehcles (seconds) L L L-10 1, L-20 2, M M-5 1, M-10 3, M-20 7, ,552 H H-5 2, H-10 5, ,147 H-20 11, ,050 9,212 *L: lo demand, M: medum demand, H: hgh demand; 1,5,10, and 20: marke peneraon rae (n percen) 4.4 Ke nsghs for V2V-based ATS As llusraed n Secons 3 and 4, a prmar benef of he graph-based approach s s capabl o rack he spaoemporal characerscs of vehcle knoledge eplcl. Furher, he proposed graph-based mul-laer neork frameork provdes an eplc modelng capabl o arculae ho nformaon flo evolves and propagaes. can lnk o he neracons h raffc flo and ner-vehcle communcaon dnamcs. The graphbased mul-laer frameork also provdes he mporan