ADAPTIVE TRAFFIC CONTROL IN METROPOLITAN AREAS

Transcription

1 ADAPTIVE TRAFFIC CONTROL IN METROPOLITAN AREAS Prof. Dr.-Ing. Bernhrd Fredrch Insttute of Trnsport Engneerng nd Plnnng (IVH), Hnnover Unversty, Appelstrße 9A, D-3167 Hnnover, Germny Dr.-Ing. Mohmed Shhn Trnsportton Engneerng Deprtment, Fculty of Engneerng, Alexndr Unversty, Alexndr 21544, Egypt 1. Context Chngng publc wreness wth respect to the mpcts of cr trffc n combnton wth the ncresng trffc demnd hs cused new orentton of trnsport nd trffc polces n metropoltn res. In order to mnmse negtve mpcts of cr trffc s pollutnt emssons nd n order to preserve moblty n sustnble wy most ctes strt mplementng strteges to reduce cr trffc whle mprovng publc trnsport t the sme tme. Due to ths focus trffc control ms to fvour publc trnsport e.g. t sgnlsed ntersectons. However, mny of those prtculr mesures do not relly cope wth the overll polcy snce these mesures nether respect mpcts they cuse n the network context nor re they ble to blnce the needs of dfferent user groups ccordng to n greed common objectve. To overcome ths unstsfctory stuton number of trffc dptve network control methods were developed n the Unted Sttes nd n prtculr n Europe nd re gettng more nd more mplemented n complex urbn rod networks (see Bblogrphy). Ths pper gves n overvew to the methodology of the trffc dptve sgnl control method BALANCE nd focuses on some prtculr nnovtve trffc nd control models tht re nvolved. In order to llow for ssessments of possble benefts of dptve control methods some results of feld studes re presented t the end. 2. Stte of the Art Wth the development of the computers n the 6es, new perspectves for the trffc sgnl control rose. The orgnl mnul optmsng methods, s for exmple the tme-dstnce dgrm for optmsng the coordnton, were trnsferred nto lgorthms tht cn utomtclly be worked by computer. On the bss of the lgorthms tht re trnsferred nto softwre, t ws ether possble to determne sgnl-schedules ftted to typcl demnd stutons nd to use them t the rght tme (offlne), or to control the trffc sgnls wth the control-softwre drectly on the computers (onlne). The frst use of modern computers for the trffc control took plce n Toronto, Cnd n 1963 (MACGOWAN et l. [1]). Shortly fter tht nstlltons of computers followed n Glsgow, Scotlnd (HOLROYD et l. [2]) nd n Sn Jose n the USA. In the 7es nd 8es n ongong development of both the hrdwre nd the control methods for computer systems took plce. Here two dfferent pproches of the trffc-dependent control re pursued: - the logc-bsed (vehcle ctuted) pproch represents the converson of expertknowledge nto logc wy. For ths pproch trffc engneers drw up logcs for the control of ndvdul ntersectons. Testng chrcterstc mesurements (e.g. the tme gp between vehcles) control decsons re derved n the logc. The result of the control corresponds to the engneer s expertse who drws up the logc. However, the qulty of control decson s not evluted by the method. Therefore n ctul optmston does not tke plce. In generl the supples of the trffc sgnl controllers convert the flow dgrms of the ndvdul ntersectons to n ech tme unque executble code usng specfc progrmmng lnguges. 1

2 - Adptve pproches re bsed on the onlne modellng of the trffc stuton nd the mpcts whch cn be quntfed wth tht. A control model descrbes the vrbles nd constrnts. The optmston systemtclly tests the possble control djustments nd evlutes them on the bss of the modelled effects correspondng to the trget functon. The clculton requrement, whch s necessry for loctng the globl optmum, ncreses exponentlly wth the number of the vrbles. MERTZ [3] demonstrtes ths correlton nd proves tht the optmsng problem corresponds to known problem 1, for whch LENSTRA nd AARTS [4] proved the NP-hrdness. The NP-hrdness leds to the fct tht generlly the globl optmum cnnot be found n rel tme. Due to ths reson, only heurstc soluton pproches cn be used. An dptve control method s formulted n generlly vld wy for dfferent ntersectons, respectvely trnsportton networks. By confgurton dt t cn be ftted to ny surroundng. Whle n Germny vehcle ctuted control methods re used lmost exclusvely up to now, n the USA nd UK for more thn 3 yers t hs been worked extensvely on the development of generlly vld strteges tht rect senstvely, therefore dptvely, towrds dfferent trffc demnds. Due to the NP-completeness, the exct soluton of the optmsng problem s not possble for rel stutons. Therefore dfferent strteges hve been chosen n order to develop rel tme-cpble methods wth the help of smplfcton. For the requrements of smplfcton there re two fundmentl possbltes, whch usully re combned: - Sptl nd temporl restrctons of the optmsng problem through consderton of solted junctons wth two optmsed sgnl groups, lmtton of horzon (for exmple 6 s), s well s the choce of longer ntervls s bsc unt (for exmple 5 s). - Decomposton of the optmston problem nto prtl problems through the seprted consderton of control vrbles (cycle tme, offset, splt), the seprted consderton of the trnsport modes nd the dfferentton of dfferent sptl levels. Aprt from these smplfctons heurstc optmsng lgorthms re often used. After flures of the frst dptve methods n the lte 6es nd 7es (for exmple the Glsgow Experment, Pldent nd After [2], UTCS 2 reserch project [1]), the second generton 3 hd brekthrough wth the SCOOT method (Splt, Cycle nd Offset Optmston Technque) n the erly 8es. SCOOT s the best-known nd wth over 2 nstlltons world wde, the most frequently used method up to now. The Brtsh Trnsport nd Rod Reserch Lbortory TRRL [5] strted to develop SCOOT n Through SCOOT, by usng the sturton level of the hghest loded ntersecton, the cycle tme wthn the network s determned nd through the sturton level of the entres the green tme t the juncton s determned. Merely the optmston of the offset s bsed on dynmc trffc model, wth whch the wtng perod cn be mnmsed consecutvely v the observton of prtl networks ( Mn-Ares ) t ll junctons. The benefts were evluted by lrge re feld trls n fve ctes n the UK nd three other ctes n the USA, respectvely Cnd (Oxnrd, Red Der, nd Toronto), n whch the method ws compred to n optmsed fxed tme control. SCOOT cheved verge dely svngs of bout 12%. The thrd generton does not use the control vrbles cycle tme, stge scheme, stge sequence, splt nd offset tht re known from the mnul plnnng, but merely consder the green tme, respectvely the red tme of the ndvdul sgnl groups s optmston vrbles. User specfctons, such s the mnmum green tme re secured by constrnts. The optmston crter, such s the delys, re determned for rollng horzon by very smple mcroscopc smulton models. Snce n ths control methods cycle tme s not used, they re lso clled cyclc methods. Aprt from the US-Amercn method OPAC [6], the French methods PRODYN [7] nd CRONOS [8], s well s the Itln UTOPIA/SPOT [9] becme known. The methods hve locl vew on juncton n common, due to the complexty of the 1 It s mtter of the optmsng problem of Sequencng weghted trdness 2 UTCS stnds for Urbn Trffc Control System 3 A dvson of the methods, whch hs been known up to the 8es, nto so clled genertons correspondng to ther cpcty chrcterstcs ws crred out by GARTNER [6]. 2

3 optmston. Whle wth dynmc progrmmng n exct optmston lgorthm s used by PRODYN, other methods use pproxmton methods 4 for resons of runnng tme requrements. Only UTOPIA/SPOT performs network-wde optmston on the bss of mcroscopc modelled network level, s well s through the consderton of the ext flowng trffc from djcent ntersectons. In contrst to the methods of the thrd generton, the coordnton n networks, s well s new demnds, such s the preference of publc trnsport, become more mportnt n the BALANCE [1] method, whch hs been developed n Germny n the 9es. Inspred by workng wth BALANCE, the further mprovement of the publc trnsport prorty n dptve control methods re n the centre of MERTZ new developments. He ntroduces trffc model for consderng the nterctons between prvte cr nd publc trnsport systems s result of ts prorty [11]. At the sme tme the vlble trffc model s bsed, smlr to the SCOOT method, on determnstc pproch. Merely the publc trnsport vehcles re showed by mcroscopc stochstcl model. The use of ths model s restrcted to one ntersecton, where n mprovement of the trffc stuton could be proved. DÜRR [12] dedctes hmself to the subject of dvdng the optmston problem nto severl prtl problems nd to the sub-optmston of the soluton whch comes together wth tht. In the DARWIN method, whch hs been developed by hm, mcroscopc smulton s ntended s trffc model for the entre network. The smulton s event-orented, n order to rech the smulton speed tht s requred for the onlne control. DÜRR fces the problem of the NPcompleteness wth the use of genetc lgorthms n the optmston. It turns out tht, becuse of the requred clcultng tme, for smll networks the pproch s not sutble n prctce. Due to the smll performnce-dvntge of the mpct-orented smulton n lrge systems, t s not to be expected tht n the ner future n pproprte pproch cn fulfl the dffcult condtons of the rel tme use. Recently, control method whch s rectng to the experences wth new pproches s ntroduced by DIAKAKI nd PAPAGEORGIOU [13] under the nme of TUC. In order to secure the control s stblty of the trffc flow tht s chrctersed by nsecurtes, TUC ws conceved s feedbck-method, of whch the modellng only bulds up on regstered dt of the precedng cycle. The entre number of the queued vehcles n the observed system, whch re mnmsed wth the help of multvrble control, s used s n optmsng crteron. Wth the use of smple model the number of jmmed vehcles s determned by mesurements of the occupncy-level t the detector. The control s represented by control mtrx, whch hs been determned offlne by lner-qudrtc optmston. The splt s optmsed (s tmecontnuous problem) t ll ntersectons of the observed system. Under consderton of the ssumptons (tme contnuty of the trffc flow, feedbck-chrcterstc), s well s of the restrctons (fxed cycle tmes nd offset tmes, no publc trnsport prorty), network-wde optml result (ccordng to the requrements of exct optmston) occurs. Aprt from the bove-mentoned new developments, the MOTION method s descrbed s dptve control [14], too. From the publctons of MOTION, t s only prtlly to be nferred tht t s mtter of model-bsed method. Except the Pth Flow Estmtor [15] for determnng the network flows, whch s known from lterture, there re no detls gven bout the models whch re the bss of the mpct determnton or bout the optmston lgorthms. Accordng to the descrpton n the bove mentoned publcton t s ssumed tht the optmston s mtter of heurstcs tht re smlr to the logc-bsed methods. For ths reson MOTION s not consdered s reference for the stte of the rt. 3. Methodologcl Approch The development of new network control methods needs to be orented on current requrements to flexblty nd comptblty. These re n prtculr 4 OPAC: Optml Constrned Sequentl Serch (OSCO); CRONOS: BOX-Algorthm; SPOT: no detls known 3

4 - Modulrty ech system component opertes even wthout centrl hrd- nd softwre nd the rchtecture of the system llows for comptble use wth dfferent other components (e.g. wth components of the exstng legcy system) - Robustness - the consequences of flure by dfferent prts of the system cn be nlysed nd protecton cn be provded to ensure contnued operton; - Evolutonrty control systems cn be extended ndependently from the orgnl confgurton; - Inter-operblty hgher performnce cn be cheved by mutul nformton s the prts of the system communcte through stndrd nterfces so tht they cn be enhnced nd upgrded wthout ffectng the operton of other prts; - Integrton the common strtegc pproch s gurnteed by the dstrbuted herrchcl system rchtecture; The system rchtecture of BALANCE tkes these ssues nto consderton by ts dstrbuted but herrchclly structured desgn (Fgure 1). Opertor Interfce Assgn dt to Network Model Network OD Mtrx Assgn dt to Sgnl Groups Impct Anlyss Optmston Reference Plns Assgn dt to Network Model Network Level Mcroscopc Dt Flow Profles Turnng Movements Assgn dt to Sgnl Groups Impct Anlyss Optmston Second by Second Control PI Locl Optmston Fgure 1: Bsc Blnce Structure At the locl level McroBALANCE rects n second by second mode to stochstc vrtons nd sudden clls for publc trnsport (pt) prorty or other events. It therefore pples smple mcroscopc trffc model nd uses rollng horzon technque to forecst vehcle rrvls. Both 4

5 prvte cr trffc nd publc trnsport vehcles re represented by ths model n n ntegrted wy, hence mpcts of pt prorty to the overll trffc stuton re drectly represented. OD estmton t the ntersecton level uses trffc flow pttern nd tme dependency to determne the respectve movements. Optmston tkes plce wthn gven frme tht constrns the locl control due to the requrements for n optml network-wde trffc stuton. At the tctcl level McroBALANCE optmses network relted control vrbles. In reference pln for ech sngle locl controller network cycle tme, offsets nd splt for dfferent stge sequences re determned. McroBALANCE uses heurstc pproch nd decomposes the optmston problem nto severl subproblems tht cn be esly solved by lner optmston. For the determnton of the relevnt crter of the objectve functon OD flows re estmted usng the nformton on movements t the ntersecton nd pplyng mcroscopc queung model. Evlutons showed tht Mrkovn chn technque usng emprclly gthered rrvl nd servce dstrbutons performs best. Due to ths rchtecture BALANCE cn be ppled t sngle juncton nd cn be extended to complete urbn network. The system cn be modfed wthout dsrupton or redundncy of exstng components - ths provdes n evolutonry pproch. The needs of network opertors to confgure the control nd to modfy t due to the dynmc spect of trffc re stsfed by n esy to understnd grphcl user nterfce. Chnges n the control strtegy cn be tested nd the effects cn be vsulsed offlne by connectng BALANCE to the mcroscopc smulton AIMSUN Objectve Functon nd Normlsed Performnce Index Adptve control methods offer the user comprehensve wy of nfluencng trffc control ccordng to the respectve polces. Therefore the objectve functon ccordng to whch the control s optmsed needs to be ccessble for the users. BALANCE provdes n open nterfce to the objectve functon tht llows to optmse due to strtegc ssues. Consequently the optmston objectve s clled strtegc system optmum nd the formule gven below show the generl pproch for ts relston. System Optmum mn! PI c q C where: PI Performnce Index sgnl groups Strtegc System Optmum sgnl groups sgnl group lnk set j c generl verge costs q trffc volume mn! PI α C where: α mtrx of concdence of lnk sets j over sgnl groups (1) (2) > α j j The pplcton of the strtegc system optmum s shown n Fgure 2. In ths exmple dfferent lnk sets (I-V) re gven tht my be controlled ccordng to rbtrry weghts. 5

6 I II III lnk set possble membershp Possble weghts for dely V IV I: mn rdl lnk - n II: mn rdl lnk - out 3 7 III: rng rod 8 IV: mn ccess - n 1 V: mn ccess - out 2 Fgure 2: Exmple for the Applcton of Weghts s prmeters n the Objectve Functon for the Strtegc System Optmum Snce ll of the consdered crter tht re used n the objectve functon re clculted by n unque trffc model t s possble to normlse the PI t the begnnng of ech optmston n order to be ndependent of bsolute vlues of the dfferent crter s well s to gurntee stble nfluence of weghts. Fgure 3 explns the pproch for the clculton of normlsed PI. opt_step nlyse mpcts clculte PI modfy vrbles ncrement opt_step n th th ck : vlue of the k crteron n the n optmston step th st reference_vluek: vlue of the weghted k crteron n the 1 optmston step f (opt_step ) /* clculte nd store reference vlues per crterum */ { for ll crter k } /* clculte PI for the n PI { reference_vlue n k j k α c th n k reference_vlue α c sg lnk set j } optmston step */ k o k ( PI k) Fgure 3: Algorthm for the Determnton of Normlsed PI s 3.2 Mcroscopc Trffc Model As mentoned bove McroBALANCE uses smple mcroscopc trffc model wth dscrete tme steps of one second. In the model ech pproch of the respectve ntersecton s consdered seprtely by lnk representng trffc between the locton of dt collecton nd the stop lne. On ths lnk vehcles re trvellng t constnt verge speed, stop t the end of the horzontl queue nd re dschrged ccordng to constnt sturton flow rte. If movement hs to 6

7 pt pt obey prorty of nother strem (e.g. left turnng movement needs to wt for n opposte strem) t wts tll the prorty movement s dschrged. For the consdered horzon of one cycle tme vehcle rrvls re mesured for the trvel tme between dt collecton nd queue end. For the rest of the cycle rollng horzon technque s used (Fgure 4). Fgure 4: Mcroscopc Trffc Model of BALANCE Publc trnsport vehcles re represented together wth cr trffc. Accordng to the methodology tht s shown n Fgure 5 the poston n queue of the pt-vehcles s clculted every second. In prllel the PI s determned from the unque objectve functon. Thus the mutul mpct of prorty schemes on publc trnsport s well s on cr trffc cn be explctly ssessed nd be consdered n the optmston n n ntegrted wy. For ths reson the user cn drectly compre the reltve weghts of pt nd cr trffc n the objectve functon nd drect nfluence on the performnce of pt s possble. pt detector length of queue: 5 veh poston of 1 st pt vehcle n queue: 3 Fgure 5: Approch for the Representton of Pt-vehcles n the Mcroscopc Trffc Model 7

8 3.3 Mcroscopc Trffc Model It could be shown tht on McroBALANCE level queue length clculton by Mrkov chn performed best. Consderng the ntervl dt between t nd t n n+1, one cn determne probbltes for the number of vehcles ( 1,2,...) n the queue t the tme t n by the vector p ( p, p, p...) n n n n 1 2 (3) nd the probbltes for the number of vehcles t the tme t n+1 by the vector p n + 1 p n + 1 p n + 1 p n + 1 (,,...) (4) 1 2 The probbltes for chngng from stte t tme t n to stte j t tme t n+1 re gven by the trnston mtrx T(dt): Tdt ( ) p... p pn... p N NN Strtng by stuton tht s determned by vector p n nd under pplcton of the trnston probbltes T(dt) the stte probbltes t tme t n+1 cn be consequentlly clculted wth n p + 1 n p T dt ( ). (6) Snce trnston probbltes chnge over tme becuse of nsttonry demnd T vres nd TT(dt,t). T(dt,t) cn be determned by Tdtt (, ) At ( ) St ( ) where: A(t): mtrx of rrvl probbltes (7) S(t): mtrx servce probbltes The mtrces of rrvl nd servce probbltes hve the followng form: (5) A( t) N N + 1. N St () s s... 1 s s s s (8) The probbltes [ ] nd [s ] cn now be formed by the results of nlytcl probblty dstrbutons or they cn be bult from mesured vlues tht re onlne gthered by the control method. Tests showed good coeffcent of determnton r 2 of.82. Dfferent to the optmston of cycle tme nd splt smple pltoon model s used for offset optmston. The bsc de of the model corresponds to the concept of SCOOT: The mpcts of the offset t the consdered ntersecton re clculted for so-clled Mn Are whch lso comprses the djcent junctons. Dfferent to SCOOT queue lengths nd delys re not determned on the bss of mesurements (.e. the cyclc flow profles) but by the help of model bsed pproch. Aprt from the reducton of dt communcton ths pproch provdes the dvntge tht no locl component (McroBALANCE) s requred. Ths comples wth the

9 requrement of modulrty nd n prtculr for lrger networks wth heterogeneous components (legcy s well s new dptve) offset optmston cn be performed. The smple pproch llows for onlne operton lso for lrge networks. Cused by the mutul mpct s consequence of sequentl optmston of the ntersectons n lgorthm hs been developed tht uses the knowledge on the OD-strems nd thus provdes systemtc dptton of offsets ccordng to the relevnce of the respectve ntersectons. 3.4 Control Model The control model of BALANCE uses cyclc pproch snce ths s the best soluton wth respect to the bove mentoned requrement of modulrty. I.e. cyclc control model llows for beng esly co-ordnted to ntersectons controlled by legcy systems lke fxed tme or vehcle ctuted control. Aprt from ths the cyclc model offers the engneer to constrn the control rbtrrly ccordng to hs needs. In ths sense BALANCE provdes dfferent mens to customse control by followng constrnts: - mnml nd mxml or fxed cycle tme - mnml nd mxml or fxed splt per stge ssgned by McroBALANCE - mnml nd mxml or fxed buffer tme per stge ssgned by McroBALANCE - fxed offset for one stge, even by chngng cycle tme nd chngng stge sequences - stge on demnd (n prtculr for pt prorty). In ths wy BALANCE control cn be constrned to fxed tme control but lso nterestng combntons my be relsed lke fxed tme control wth publc trnsport prorty s t s fvoured n Leeds [16]. However, f full demnd responsveness nd therefore full flexblty of control s desred BALANCE wll behve cyclclly becuse of the full vrblty of the followng vrbles: - cycle tme, offset, nd splt (mnmum optmston cycle s one cycle tme) - buffer tmes nd stge sequences (optmsed ech second) Usng cyclc control method for controllng network, common cycle tme or multple of t s pproprte for the locl trffc forecst nd for the co-ordnton of neghbourng junctons. Therefore the control model n BALANCE s usng common cycle tme tht s clculted due to the trffc condtons t the crtcl juncton. The decsve juncton s determned by mpct crter tht result from the clculton of the optmsed green perods for the sngle sgnl groups. Aprt from the totl dely W node t the juncton s crteron, the verge ndvdul dely w node over the optmston perod s used s well. These delys cn be weghted by the fctors δ, γ. The decsve or crtcl juncton s the juncton whose performnce ndex PI CycleTme W + γ w δ (9) node s mxml. Ths crteron ncludes the condtons n the whole network W node j node PT CT α W + β W (1) where:, j: α, β : PT, CT: ndex of sgnl groups resp. sets of sectons weght fctors used n the objectve functon of the optmston publc trnsport, cr trffc s the number of vehcles n the network s mplctly ncluded s well s the degree of mpct of the consdered sngle rod users bsed on the verge ndvdul dely w node j PT CT 1 α w + w. (11) β 9

10 By ncludng the verge dely optmston tkes cre of those rod users t smller junctons (wth less numbers of lnes per entry secton) f there s hgher dely. For the decsve juncton cycle tme wll be optmsed tkng nto ccount the constrnts of the defned green perod proportons nd usng mcroscopc mpct model. The objectve functon n tht cse s focused on the mxmston of the cpcty whch s equl to the mnmston of the dely: mn PI( tcyc) j CT α W ( tcyc) (12) tcyc tcyc tcyc where: tcyc: cycle tme; mn mx By the pproch of mxmsng cpcty for cr trffc t s ssumed tht publc trnsport vehcles tht don't hve segregted lne wll proft from the optmston n the sme wy s cr trffc. Hvng clculted cycle tme McroBALANCE optmses the splt. By the mnmston of lner combnton of the totl nd verge dely W CT nd w CT respectvely: mn PI( tg ) β j CT CT α W ( tg ) + w ( tg ), (13) where: tg : green tme of phse ; tg mn tg tg The results of the optmston form frme sgnl plns for ll ntersectons of the consdered network (Fgure 6). These frme sgnl plns gve the possblty to rect to short term fluctutons n trffc demnd by McroBALANCE especlly on those junctons tht re less sturted. The buffer tmes tht cn be used depend drectly on the predefned frme vlues (mnmum nd mxmum green tme Mn x nd Mx x ) nd constrnts on the mnml tmngs MI x. mx MI1 K2 Mn1 MA1 K2 Mn1 MA1 Mn2 MA2 K1 Mn2 MA2 MI2 MI2 t c : Mn x : MA x : K x : MI x : Cycle tme Mnmum green tme of stge x. For ths durton green hs to be dsplyed ndependently from trffc demnd. Vrble (trffc demnd dependent) prt of green tme tht s ssgned by McroBALANCE; rnge [, t c - ΣMn x ]. Kernel green tme of stge x. Resultng green tme of stge x fter optmston by McroBALANCE; ΣK x t c. Vrble prt of MA x (buffer tme) tht s ssgned by McroBALANCE due to short term vrtons n trffc demnd; rnge [, MA x ]. Fgure 6: Defnng Frme Sgnl Pln nd dstrbutng Green Tme to Stges t c McroBALANCE provdes to swtch smoothly between severl stge sequences (Fgure 7). In ths wy not ll stges need to be dsplyed ech cycle tme nd thus n cyclc control s fesble wthn cyclc frme. In order to serve prtculr demnds (e.g. clls for selectve vehcle prorty) n mmedte chnge to the pproprte stge sequence s ppled. Dynmc dptton of stge durton llows for copng wth selectble co-ordnton fxponts. 1

11 Coordntonfxpont Decson pont for swtchng to b Decson pont for swtchng to Sequence Sequence b Proportonl dptton of stges Perod of swtchng from to b Fgure 7: Dynmc Chnge of Stges n BALANCE 4. Results of Feld Trls nd Smulton Studes A frst verson of BALANCE hs been developed n the frme of the Europen R+D project Munch COMFORT nd hs been tested nd ssessed n feld trl s well s n smulton study. The prototype developed n Munch COMFORT hs then been further extended nd mproved n the frme of the EC funded demonstrton project TABASCO. In prtculr new verson of BALANCE tlored ccordng to the hgh requrements of UK control trdtons hs been developed nd ws mplemented nd trled n London nd Glsgow. All trls nd studes performed n the frme of the mentoned projects show comprble results: - Sttstclly sgnfcnt on the 95%-level nd wth confdence ntervl of 3.5% the smulton study for the Munch trl ste 5 wthn Munch-COMFORT showed tht usng only the tctcl level dely of cr trffc could be reduced by 19% provdng the sme qulty of pt prorty. - A nne dy feld trl for the Munch test ste resulted n comprble fgure: The ssessment showed tht 14% of the cr trffc dely cn be sved enhncng the exstng v-technque by McroBALANCE (Fgure 8) dely of cr trffc dely of pt Exstng v control method wth pt-prorty BALANCE Fgure 8: Reltve chnges n dely t the Munch trl ste Dchuer / Pelkoven Strße - The soco-economc evluton whch ws performed n the frme of TABASCO ccordng the Germn gudelne EWS 6 resulted n cost/beneft rto of 1/ Trl ste ws the ntersecton Pelkoven Strße / Dchuer Strße EWS: Empfehlungen für Wrtschftlchketsuntersuchungen von Strßen 11

12 - In the frme of the London trl BALANCE ws compred to the dptve control method MOVA [17]. In ths trl both BALANCE levels Mcro- nd McroBALANCE were tested. In spte of dffcultes wth respect to the onlne dt detecton good results could be cheved (Fgure 9): - The verge dely of cr trffc could be reduced by 4% - The longest dely of vehcle ws shorter by 11% - The stndrd devton of dely ws reduced by 45% Dstrbuton of prvte trffc dely MOVA BALANCE s s Fgure 9: Reltve chnges of dely nd the respectve stndrd devton t n ntersecton n London/Hethrow [18] - In Glsgow prtculrly the ntegrted pt prorty functon hs been tested. In ths trl mprovements of the pt trvel tmes of 7% could be shown (Fgure 1). 2:9 1:55 1:4 1:26 1:12 :57 :43 Before After After After Before After Before After Mx Av. Mn :28 :14 : Before Before Tme Perod Fgure 1: Reltve chnges n trvel tme of pt vehcles nd the respectve spn wdth for the trl n Glsgow [18] 5. Summry nd Perspectves Aprt from sfety relevnt spects trffc sgnl control n metropoltn rod networks needs to consder n prtculr requrements wth respect to economc operton nd urbn plnnng polces. Trdtonl control methods tht only regrd the locl envronment nd tht decde on the bss of smple plusblty crter do not fulfl ths requrement. However, the developments 12

13 MESSE- HAUS n trffc modellng enble to quntfy onlne the mpcts of control nd to reflect them n n objectve functon to be optmsed. Accordng to ths m nd wth respect to precedent work n ths re the rchtecture for new dptve sgnl control method hs been developed. The open rchtecture provdes the frme for the jont operton of dfferent components,.e. of dvnced dptve nd legcy modules. Thus the requrements concernng flexblty nd comptblty re fulflled (.e. s mentoned t the begnnng: modulrty, robustness, evolutonrty, nter-operblty nd ntegrted operton). The system rchtecture dstrbutes the control problem on three levels: The strtegc level offers the opportunty to ctvely nfluence control ccordng to trffc polces (strtegc system optmum). On the tctcl level the trffc stuton n the consdered network s modelled onlne nd thus llows for the optmston of network-wde relevnt control vrbles tht form frme plns for the locl control. Wthn the frme plns the locl control rects to stochstc vrtons due to the results of onlne mcroscopc modellng. The overll system s relsed by the commercl softwre BALANCE. Evlutons n the context of dfferent demonstrton projects (TABASCO [18] nd others) hve shown tht the pplcton of BALANCE sgnfcntly reduces costs n terms of dely nd queue lengths. In verge svngs of bout 15% compred to vehcle ctuted control could be cheved. Lrger scle mplementtons n Munch (Fgure 11) nd other ctes re currently n progress. Fvoured by ts system rchtecture BALANCE wll become mn buldng block of strtegc control system n Munch tht llows for ctve nfluence on dfferent control methods nd networks by supervsor. Trffc Adptve Control t Juncton Steuerungsgebet BALANCE n Rem Verkehrsdptve Knotenpunktssteuerung Detectors Detektoren TRANSVER 1/99 94 Schtzbogen Sthlgruber - Rng Remer Str. AS Am Moosfeld Mttbcher Str. Grf - Lehndorff - Strße Am Mtterfeld Lebengerstrße Remer Str. AS München - Rem 94 Am Helgrben FEUER- WACHE An der Pont Olof - Plme - Str. SEE P SEE U ICM TAGUNG U2 Messestdt West P 94 AUSSTELLUNGSHALLEN Pul - Henr - Spok - Strße P Wlly - Brndt - Allee Hofbräu - Allee AS Feldkrchen West P P+R U U2 Messestdt Ost Alfons - Goppel - Strße P Ottendchler Str. De - Gsper - Bogen Fgure 11: Overvew to the Rod Network n the re of the new Munch Trde Fr wth more thn 2 BALANCE controlled Intersectons 6. References [1] MACGOWAN, J.; FULLERTON, I.J. ( ): Development nd Testng of Advnced Control Strteges n the Urbn Trffc Control System, n: Publc Rods, Vol. 43 [2] HOLROYD, J.; HILLIER, J.A. (1971): The Glsgow Experment: PLIDENT nd After, RRL Report 384 [3] MERTZ, J. (21): En mkroskopsches Verfhren zur verkehrsdptven Knotenpunktsteuerung mt Vorrng des öffentlchen Verkehrs, Schrftenrehe Fchgebet Verkehrstechnk und Verkehrsplnung, TU München, Germny [4] LENSTRA, J.K.; AARTS, E.L. (1997): Locl Serch n Combntorcl Optmzton, Chchester [5] HUNT, P.B.; ROBERTSON, D.I.; BRETHERTON, R.D., WINTON, R.I. (1981): SCOOT - trffc responsve method of coordntng sgnls, TRRL Lbortory Report 114 [6] GARTNER, N.H. (1982): Demnd-responsve decentrlsed urbn trffc control, US Deprtment of Trnsportton DOT/RSPA/DPB -5/81/24, Wshngton

14 [7] HENRY, J.J.; FARGES, J.L.; TUFAL, J. (1983): The PRODYN rel tme trffc lgorthm, Vortrg m Rhmen des IFAC Symposums, Bden [8] BOILLOT F. (1992): Optml Sgnl Control of Urbn Trffc Networks, Proc. Of the 6th Int. Conf. On Rod Trffc Montorng nd Control, IEE London, no 355, pp [9] DONATI, F.; MAURO, V.; RONCOLINI, G.; VALLAURI, M. (1984): A Herrchcl Decentrlsed Trffc Lght Control System. The Frst Relston: Progeeeetto Torno, IFAC 9th Worl Congress, Vol II, 11G/A-1 [1] FRIEDRICH B. (1999): En verkehrsdptves Verfhren zur Steuerung von Lchtsgnlnlgen, Veröffentlchungen des Fchgebets Verkehrstechnk und Verkehrsplnung der Technschen Unverstät München, München. [11] MERTZ, J.; WEICHENMEIER, F. (22): Modellbserte multmodle LSA-Steuerung n Echtzet, Vortrg m Rhmen der Heurek 22, Optmerung n Trnsport und Verkehr, Krlsruhe [12] DÜRR, P. (21): Integrton des ÖPNV n de dynmsche Fhrwegsteuerung des Strßenverkehrs Steuerungsverfhren DARVIN, Heft 11 der Veröffentlchungen des Lehrstuhls für Verkehrs- und Stdtplnung, Technsche Unverstät München, München [13] PAPAGEORGIOU, M.; DIAKAKI, C. (1999): Applcton nd Evluton of the Integrted Trffc-Responsve Urbn Corrdor Control Strtegy IN-TUC n Glsgow, Vortrg m Rhmen des TRB Annul Meetng, Wshngton [14] KRUSE, G.; BUSCH, F. (22): MOTION for SITRAFFIC Optmerung der Lchtsgnlsteuerung m Enstz, Vortrg m Rhmen der Heurek 22, Optmerung n Trnsport und Verkehr, Krlsruhe [15] BELL, M.; LAM W. ; IIDA, Y. (1996): A tme-dependent multclss pth flow estmtor, Vortrg m Rhmen des 13th Interntonl Symposum on Trnsportton nd Trffc Theory, Lyon [16] HALLWORTH, M.S.; JONES, S.E. (2): Selectve Vehcle Prorty A New Approch. Proc. of the Tenth Interntonl Conference on Rod Trnsport Informton nd Control 4 6 Aprl 2, Publcton No. 472 IEE, London, pp [17] MAURO, V.; DI TARRANTO, C. (1989): UTOPIA. Proc. of the IFAC Symposum, Prs [18] TABASCO REPORT 9.3 (1998): UTC wth PT Prorty: User Needs, Functonl Specfcton nd Demonstrtors (Ed.: Fredrch B.). Project number TR154 Europen Commsson, DG XIII, Brussels 14