A HIERARCHICAL APPROACH TO MODELLING NETWORK FLOWS

M"A.P. TAYLORSenior Research ScientistCSIRO Division of Building ResearchMelbourne

M" ANDERSONExperimental OfficerCSIRO Division of Building ResearchMelbourne

ABSTRACT: The size and oompLexity of' urban transport systems requirethat modeLs o.f suoh systems be designed to operate IJithinpmct'iaat and manageabZe Zimit;s on da:ta roequiroements andcomputationaL effort.

One apP1'oach -is to use a hier>ar>ahy ol troanspoT't flOl.J model.s,in which a set of system level-sis defined, and parotiautaf'modeLs used for anaLysis at each LeveL.

This paper desoribes the de.finiticm of a hierarchy o.f networkf7,oo models, and the "Z,assifieation of vanOUB modets withinthis hierarohy. An inte(lrated set of' modeLs IJithin thehiem;roahicaZ f1'arnBW01'k is descnbed. This set aonBists of' someestablished modeZs together' with 80me moroe roe-cent modeZdeveLopment. The LeveLs within the hierarohy ran(Je frommicroo-·Zevel, jlOlJ) sirrruZation th1'ough z.oeal. al"ea ne1;Wor'ks toZaroge-s(!.aZe ne-tl.uoroks, and stl'lategia Zand use-t'Y'anspo'Y'tinte1'act·ion modete. Exampl.es of the use ol modeZs .for' eachare (liven, and the methods jor Unkin(l modeLs are defined andempLored.

136

INTR~

modeand I

theof ~

segnandbe l

unalto (enelill 1

enCi

repandtrabeetesSAE19Eth2tf'cimane

CO!syurlstabtrindeinGiccVEoi

s~

elrnlp'i(apa

rr

( FLOWS

rtems -pequi7'8

'pate uJi thin'ements and

·t ftObJ modets.Id paptieu'La1'

t1'chy of netl.>o1'k, modets withinlithin the~ consists of,ent modetflange f1"om

I netwo1"ks totpansp0r't?dets fo1' eacha1'e defined and

MOOELLING NETWORK FLOWS

INTROOUCTlON

The size and complexity of urban transport systems require thatmodels of such systems should be designed to operate within practicaland manageable limits on data requirements and computational effort. Atthe same time, problems often arise concerning the detailed performanceof some components within a network (e.g .. a specified link, node, orsegment). Possible impacts of policy and operational changes at localand system-wide levels may be of interest, and modelling procedures maybe useful tools for these problems. However, individual models may beunable to provide the full range of possibilities and levels of detailto assist in an analysis of this type .. Recent interest in modellingenergy consumption and emissions in urban traffic systems proVides anilluminating example of this type of problem and the difficultiesencountered ..

Considerable work in research and policy formulation has beenreported on the overall consumption of energy by transport (e .. g .. Lawlorand Brown 1980) and the overall emissions of pollutants fr.om roa.dtransport (e,9. Environment Rrotection Authority, 1979). There has alsobeen much interest in the performance of individual vehicles on thetest-bed, or in traffic, as shown in the proceedings of the two jointSAEjARRB conferences on traffic, energy and emissions held in 1980 and1982 (ARRB 1980, 1982) .. However the middle ground between these areas,that of the performance of traffic systems and their component (mixed)

streams, has been neglected. It would appear most useful toa methodology for connecting these areas of policy interest

research endeavour within a unified framework ..

Urban traffic is an example of a large-scale system involvingex interactions between its components, while itself forming a sub­

within higher-level systems (the urban transport system, and theactivity system) .. The interactions between vehicles, traffic

controllers, components of other transport sub-systems, andland use activities mean that the overall performance of the

is not a simple function of the performance of itscomponents. An interesting example of such effects was

ibed by Gipps (1981) concerning the performance of a traffic streamwhich some petrol-driven vehicles were replaced by electric vehicles.

found that for the arterial road studied, the total fuelconlsurnption of the traffic stream could be increased unless the electric

had acceleration and cruising characteristics simil ar to thoserest of the traffic.

One approach is to use a hi erarchy of model s, in whi ch a set oflevels are defined, and particular models used for analysis at

leveL If connections for information transfer between levels ands can be establ ished, then an overall systems analysis can be

per'fo'f"""" The effects of system or component changes may beover a range of levels of aggregation Studies at the low

evel of the hierarchy may be used to produce submodel s for uselevels. Conversely higher level network models and land use

model s may be used to gener:"te travel demand data for localindividual link models.

This paper defines a particular hierarchy of network flowand classifies a number of models within this hierarchy .. An

137

TAYLOR AND ANDERSON

integrated set of models within the hierarchical framework is described,consisting of some established models together with some more recentmodel developments. The hierarchy presented here is a development fromthat suggested by Taylor and Gi pps (1982). Changi ng data requi rementsand constraints for different model levels within the hierarchy areillustrated, and methods for linking particular model levels areindicated.

A HIERARCHY OF NETWORK FLOW MODELS

The following hierarchy is proposed for transport networksystems analysis, from the most detailed (micro"level) to the regional(macro"level) :(a) microscopic simulation of individual units in a traffic stream.

This level includes models such as the multi"lane arterial roadsimulation model MULTSIM (Gipps and Wilson 1980), the rural roadsimulation mOdel TRARR (Hoban and McLean 1982), and the simulationof pedestrian movements in a plaza (Sands and Ciolek 1979);

(b) macroscopic flow modelS in which the flow units ar'e assumed tobehave in some collective fashion. Examples of such modelS includethe kinetic flow model for a length of road proposed by Herman andPriogine (1979), flow"travel time models (e"g .. Davidson 1966), thequeue dynamics models for signalised intersections (Stephanopoulos,Michalopoulos and Stephanopoulos 1979), platoon dispersion models(e,,9. Seddon 1972), and pedestrian flow models (e.g .. Fruin 1971);

lc) simulation models of flows in networks for the optimisation ofnetwor k per formance for fixed route choi ce preferences" TRANSYT(Vincent, Mitchell and Robertson 1980) is the best known model ofthis type;

ld) local area network assignment models such as LATM (Taylor 1979),SATURN (Hall, Van Vliet and Willumsen 1980) andCONTRAM (Leonardand Tough 1979), wh i ch model route choi ce throUgh detailed no,ewo,'representing small parts of an urban area" Also included arepedestr i an network flow model s such as that devi sed by Ceder(1970) ;

le) large"scale network assignment models such as TRAFIC (Nguyen 1974)or ARRBTRAFIC (Luk, Markarov and Wigan 1979) which may includelarge sections of an urban area (e"g" Taylor and Anderson 1982) ora complete metropolitan area (e,.g. Boyce, Jansonand Eash 1981);

(f) large"scale sketch planning models of land use/tranSportinteractions such as TOPAZ (Brotchie, Dickey and Sharpe 1980),which estimate flows between activity centres without explicitlyconsidering the transport network; and

(g) models of regional, national and international flows of people andgoods (e .. g Batten 1982) ..

This hierarchy should not be seen as completely definitive, andthere are some overlaps between adjacent levels The possibil itiesexist to distinguish levels on the basis of 'study area' or 'studyduration', with both the area and the duration typically increasing asone moves from level (a) through to level (g)" Study areas increasefrom a single length of road, to areas of several hectares, to severalsquare kilometres, to entire metropol itan areas, and beyond .. Studyperiod durations for microscopic simulations would generally be of theorder of an hour or less, increasing to a peak hour (or hours) for thelocal area models to 24 hour periods for the sketch planning models andperhaps months (or years) for national/international flow models. Ananalogy might be drawn with the use of a microscope., The lower levels

138

obf

C

sdvt

o

h

,work is described,lme more recentI development fromlata requirementshierar'chy arelevels are

port network) to the regional

a tr affi c stream.lane arterial road80), the rural road), and the simulCio1ek 1979);

ts are assumed toof such models incl,roposed by HermanI. Davidson 1966),:tions (Stephanopou1,on dispersion models (e.g,. Fruin 1971)

le optimisation of'eferences, TRANSYT, best known model

LATM (TaylorlOd CONTRAMrough detail ed~l so incl uded aredevi sed by Ceder

s TRAFlC (Nguyenwhich may includeand Anderson 1982)

nson and Eash 1981)se/transportand Sharpe 1980)

s without explici

:ely definitive,,e possibil itieslrealor 'study:ally increasing~y areas increase~ctares, toj beyond,lenerally be(or hours) forplanning models

1 flow models. AnThe lower 1eve1

MODELLING NETWORK FLOWS

of the hierarchy allow a small area to be studied at high magnification,but as the area of study is increased by reducing the magnification, thefiner detail s disappear.

Constraints on Applying Individual Models

The need for a hierarchical approach to transport networkmodelling arises from the particular phenomena to be studied, andthe data resources and computational effort required to model thesephenomena. Significant changes in perspective and planning objectivesoccur as one moves through the hierarchy. At the lower levels, studiesmight be required of the effects of traffic management policies, In themiddle range of the hierarchy, the planning, location and design ofindividual traffic generators (e .. g. shopping centres) are of interest,while at the higher levels broad land use and transport planningpolicies, and economic policies, form the objectives

By way of example, consider a road traffic system. At themicroscopic level, detailed study of vehicle progression along a road isof interest, including lane choice, interactions between followingvehicles, time spent in queues, and overtaking .. At level (b), themacroscopic link flow level, mean travel time, delays and queue len9thson the link are typical dependent variables. Models at level (c) arealso concerned with these variables, but also attempt to optimise themacross a network. The route choice problem emerges at level (d), at

ch level network links can still be directly identified withindividual roads. Questions arise concerning detailed route choice

,diversions around 'bottlenecks' and the use of 'rat runs') as aof queuing points, delays and turning manoeuvres at

intersections, and the split of traffic between major and local streets,higher level models (say levels (e) and (f» are concerned with

ems of mode, destination and route chOice, plus overall networkand congestion characteristics .. At these levels network links

rp[,re'spnt. only the important parts of the road system (e,9. freeways andarterial roads), and if the network is diffuse, may use transport

'corridors' to represent groups of parallel roads,

The data requirements for the lower level models (e,.g ..fication of lane configurations, signal phasings and settings,s of road geometry and alignment, vehicle fleet composition and

"ohie'o performance characteristics) make it impractical to analyse morea relatively short length of road (perhaps up to 3 km) Computer

,,,''',,'','" time also limits the scope of a simulation run to a reasonableof vehicles (perhaps 2000 vehicles for a simulated time period of

minutes or less). At higher levels, the specification of thelength of road in a network demands less detail for eachroad link. At the local area level (level (d», individual

may still be identified in broad terms (e. g,. road type, number of, intersection types) but specific details (e.g. regarding ali9nment)

omitted At the metropolitan-wide level (level (e» even fewers can be given, perhaps only broad identifiers of road type and

land use classification. At level (f) only parameters of1 network performance are used, such as mean travel speeds ..

Continuing work within the CS1RO Division of Building Researchconcerned with the definition and development of various

139

--

Fl

a9

bet,theofi nflupPior';'pertraproandinfbeandexp

direpeday'ecapeperireqladdicharmodEby 1

fAYLOR AND ANDERSDN

Figure L OPUS Network Flow Model s System140

Linkages Within the Opus System

The interrelationships between these models can be seen inFigure 1 which depicts their relative positions in the hierarchy and the(two-way) linkages between them. Also indicated is the nature of these1inkages i.e. from the top to the bottom, outputs from one modelrepresent inputs (or constraints) to the next. The progression can beviewed as levels of demand information .. Output from the upper levelmodel might be a trip demand origin-destination matrix for' a large scaleurban road network, which would be used as input to the next lower levelto yield traffic demands for all roads in a local area network, andfinally to the lowest level where the demand would be an estimate offlow on one particular road ..

computer based transport network flow models, each of which can beidentified with various levels of the hierarchy of models describedabove.. These models are being developed to common standards and arecollectively grouped into an urban planning model package called OPUS ­Optimal Planning of Urban Systems.

Principal models of the OPUS package include, the multi-lanearterial road simulation model MULTSIM (Gipps and Wilson 1980, Gipps 1982),models concerned with road networks from a local area (suburb) scale(LATM, Local Area Traffic Model, Taylor 1979, 1982a) to the metropolitanscale (OPERA, OPus Equilibrium tRaffic Assi9nment model, Taylor 1982b)and a comprehensive land use and transport planning model TOPAZ82(Technique for Optimal Placement of Activities into Zones, Sharpe,Wil son and Pall ot, 1982).. OPUS incl udes additional model s for energyconsumption and pollution emission and dispersion from road networks,URPOL (URban POLlution model, Tay10r 1982c) and POLDIF (POLlutionDIFfusion model, Anderson 1983, Taylor and Anderson 1982) ..

ME~N FLD\.[

R'lE, AND ROAD

S£CTION TR~VEL

TlHE,

RO'U SEO,ON TR,vEL TIMES

RELAl <ON5 BETwEEN FLOV

lR~YEL TIMES ENER(~

liNK ,LOWS

TR~vEL COSTS

E..EIl6Y M.j()

EMISSION,

TO

rOPAZ82 OPERA LA rl1 II1ACRO- f1ULTSIf1-- FLOW

, ,<'>('-7:: NEWORK TRIP

EMND CvERAL

rOPAZ82 SV~':1-0 lIE/.!< T~'.vEL

<;-', COSTS

,ONGESTION SUS-'REA MEAN L'''''lEIIHS TRIF m'F' OEN'NOS

OPERA COSTS. TOTAlTRIP SEGMENT

FLOW A~IES ~ND

~""RGY ANDT~'V£L COSTS

EH'55IQNS COHS

LArl1

I1UL r5111

I1ACRO­FLOW

e-----

2. Information Transfers Within OPUS Network Flow Models System

MODELLING NETWORK FLOWS

Reverse linkages or feedback loops operate in the oppositedirection where model outputs are interpreted as indicators ofperformance of the road system, i. e., system response information Theseare measures of the adequacY of ~lz of the individual road or networkcapacities to cater for the expressed demand, A failure or poorperformance of some system response indicator at a lower level wouldrequire a reappraisal of upper level modelling efforts with changed oradditional input constraints. An example here would be a necessarychange to the allocation of land uses in an urban area by the TOPAZ82model if the OPERA and URPOL models found that pollution levels emittedby traffic assigned in the original allocation were unacceptable.

More specific indications of the types of information flowingbetween levels of the modelling hierarchy are shown in Figure 2, Herethe demand information flowing down the hierarchy is shown to the rightof the main diagonal, and the system response or adequacy of supplyinformation flowing up the hierarchy is shown to the left, Thus at tneupper levels we have the output from TOPAZ82 in the form of, say, anorigin-destination trip demand matrix for a large-scale network,perhaps comprised only of travel corridors, together with overall meantravel costs on the network being supplied to OPERA In turn OPERAproduces sytem response indicators of levels of congestion, aggregateand mean trip costs and energy consumption and pollution emissioninformation for the same network. Areas of poor system performance canbe identified and fedback to TOPAZ82 to revise the trip demand matrixand/or the land use allocation plan,. Information measures here areexpressed in either gross (aggregate) terms or as mean values.

141

At the lower hierarchy levels the emphasis shifts away from,aayp,,",p measures towards detailed information for individual roads or

h can bedescribedds and areca11 ed OPUS -

mul ti-lane980, Gippsurb) scalee metropolitanaylor 1982b)TOPAZ82

Sharpe,for ener9Y

d networks,Ll ution

seen in'Yarchy and theture of thesemodel

ssioncan bepper level

a large scalext lower levelWOY'k, andstimate of

EXAMPLES OF MODEL LINKAGES

MULTSIM to Macro-Flow MOdels

with sconcerprav; (

tractil

est i mashoe 5

Prigogin whidifferFurthein equdistrithe vaunity,betwee

road dlvol umeas a weasilyphotogthe refracti

and fo

wher e I

and m,facil i

where ''goadnEand lt~

and Medata p'all vel

(R2 varespec5% 1ev

(1)142

To illustrate the roles of the various models in the hierarchyand the relationships between them several examples of model use can bepresented ..

TAYLOR AND ANDERSON

sections of a road Thus from LATM or macro-flow type models, meantraffic flow rates and travel times for individual roads for specifictimes of day, say, a morning peak hour, are available. MULTSIM can thenmake use of this information to determine the response of individualvehicles on such a road under the given demand conditions, and outputdetailed data on individual vehicle travel times, fuel consumption,queuing delay, etc.

At the middle levels of the hierarchy, information flows areconcerned with both local area networks and individual roads, but arenot concerned with individual vehicles OPERA can be used to pr-oduce alocal area trip demand matrix to be used as input to LATM which, in turn,would provide detailed flow and travel cost information for links andintersections within the ar·ea. Both OPERA and l.ATM require expressionsindicating how travel costs (or times), energy and emission rates varyon individual road types according to the level of congestion or trafficflow on the road. These expressions can be obtained from macro-flowtype models.

The process of feedback ot information from lower levels to higherlevels and of the varying emphasis of the models can be appreciated byexamining the means of calculating tuel consumption in each of the models.MULTSIM uses a very detailed fuel consumption sub-model for individualvehicles derived from Kent, Post and Tomlin's (1982) work on enginemapping. It requires instantaneous estimates of a vehicle's speed andacceleration, updated more than once per second of travel time. Whenaggregated over all the vehicles on a road it provides a figure whichcan be compared with that from a suitable fuel consumption sUb-model forLATM such as the elemental fuel model (Akcelik, Richardson and Watson1982) which takes into account vehicles' cruise times and number of stopsper road link.

Aggregating over road links yields values to be compared withthose of the next hierarchy level model, OPERA, which relates fuelconsumed only to mean link travel times ignoring queuing and delayeffects at intersections.. At the upper level, TOPAZ82 may not evenconsider vehicle speeds, but simply use some fleet averaged value of1itres consumed per 100 km of travel.

Macro-flow model s are concerned with the aggregate behaviour ofvehicles on a road whereas MULTSIM is concerned with the behaviour ofindividual vehicles. AnderSon (1981) described the use of ~lULTSII1 tomodel individuil1 vehicles travelling on a multi-lane arterial road andthe aggregation of their behaviour to calibrate a macro-flow model

Herman and Prigogine (1979) postulated a macro-flow model wherethe average (space mean) speed of moving vehicles, vr, in a trafficstream is related to the proportion of stopping vehicles, Fs , at anyinstant by,

v = v (l_F)nr m S

MOOELLING NETWORK FLOWS

Herman and Prigogine also postulated a relatlOnship between stoptraction, Fs ' and concentration k, viz,

Fs = (k/kj)m (3)

where vm is the average maximum running speed and n is a measure of the'goodness ' of the traffic system, i "e" the standard of the road systemand lts traffic control devices" Data from roads in London, the U"SA,.and Melbourne indicate that n takes values between 1 and 3,. The Melbournedata plotted very closely by slightly above n=l. The average speed ofall vehicles, v, is related to the average speed of moving vehicles by.

(2)

(4)

WhlCh tratfic jams),of quality at the road

143

v = v (I-F)r s

v = 597 (I-F )1.37r s

for the stop fracti on-concentr ati on expr'ess i on,

F = 0,94 (k/k ,)084 (5)s )

of the log transforms at these equations were 0.90 and 0,.89y, and all coefficients were signficant at better than the

where k/" is the average maximum concentr'ation (atand rn, ike n in equation (1), is another measurefacilities and traffic control system.

MULTSIM was set up to model a 2 25 km length of a Sydney arterialroad during peak hour and to record speeds for individual vehicles andvolume. concentration and stop fraction characteristics for the roadas a whole" Data covering a large range of Fs values (0 to 0.9) wereeasily obtained compared with the laborious method of analysing aerial

suggested by Herman and Prigogine, Regression analyses onof the simulations yielded the following for the speed-stop

fraction relation,

The value of n (1 37) in equation (4) was within Herman andigogine1s expected range of 1 to 3 and was close -to the end of the rangewhich earlier Melbourne data were placed, perhaps indicating a general

between Australian traffic and that of Europe or America,,,",tho,, the value of vm' the average maximum running speed of 59 7 kmjh

on (4) was close to the expected value, the mean of theon of drivers' desired speeds, 60 km/h In equation (5),

value of the coefficient (0.94) was close to the expected value ofsuggesting that Herman and Prigogine1s postulated relationship

bei:we"n stop fraction and concentration holds true

The combination of equations (1), (2), and (3) with coefficientfrom (4) and (5) yielded an expression tor the classic horse­

shaped curve of the speed-volume relationshi-p. viz,

q = kv • k, v [1 - (v/v )1/n+1Jl/m (6)J m

similar expressions for the speed-concentration and volume­relationships" Curves tor' these three expressions,

an excellent fit to the simulation with the curves well

(1 )

the hierarchy>del use can be

ldels, meanfor specific

IULTSIM can then, individual;, and outputmsumption,

;e behaviour ofbehaviour of

)f MULTSm to~r ia1 road andFlow model.

low mcdel where1 a trafficI Fs , at any

:ompared wi thates fueland del ay

Ly not evenled value of

levels tolppreciated bylch of the models'or individual: on enginee1s speed and

time. Whenfigure which

In sub-model forIn and WatsonI number of stops

In flows arelads, but are,d to produce a1 which, in turn,'or 1inks andire expressionsion rates vary:tion or traffic, macro-flow

144

pARAMET

(a) staMel

Co (mi n

J

s (veh/

Si9nal iinter sedensity(intn,';

Fi gur'e

types 0'

'Middl elane USldensityfor theinter'se 'shown i I

Tay10r

2000

VOLUME (2 LRNES) VEH/H1500

10

:i60

SPEED l-~~-'" ,'"" I~)0 50 ' .w ,LJ '"'-Q..lil '''< VOLUME

40,

*lO

Figure 3, Ideal ised form and Herman and Prigogine' s (1979) TheoreticalSpeed-Volume Curve SuperimpoSed on Simulated Data Points

Fe (L/lOD km) = 664 + 3,59 c

Macro-flow models may be used to generate simple expressionsrelating travel costs, fuel consumption and pollutant emission rates totraffic flows for broad identifiers of road types" Such expressions arerequired by trip assignment models such as LATM and OPERA, For examplethe foilowing relation between mean fuel consumption (FC) and unittravel time c (min/km)

20

for the 1978 NSW private vehicle fleet, was derived for use in OPERA(fURPOL) by Taylor and Anderson (1982) from data given by Kent (19~O)

c = Co (1 + Jq/\s-q)) (8)

where q (veh/h) is a I ink traffic flow, c (min/kmJ is a unit traveltime, and c , J and s are parameters depending on link type andsurroundin90environment, Separate relationships were estimated for three

For a recent study of pollutant emissions over the r,lelbourneITietropol itan area usin9 DPERA/URPOL, relationships between traffic flowand travel time on various classes of road were needed" In theparticular case of four-lane undivided arterial roads, the data andmacro-flow relationships given by 8eard and ~1cLean (1974) were used tocalibrate Davidson congestion functions (Davidson 1966)

TAYLDR AND ANDERSON

representing their idealised forms As an example, the speed-volumecurve is shown in Figure 3,

f!acro-flow to r,liddl e Level Model s

OUTER SUBURB

0,.80

o 4861741

024(0 ,.15)

085

0,,468

1675052

(0,07)

OUTERSUBURBS

R 0 ADC L ASS

MIDDLE SUBURB

INNERA~A

145

INNER AREA

1 150475

1394L97

(0,,34)

Flow-Travel Time Relationships for Undivided Arterial Roads inMelbourne

standard error ;n parenthesis, based on sample counts from maps inMelway (1979),

TABLE 1DAVIDSON FUNCTION PARAMETERS FOR THREE CLASSES OF

4-LANE UNDIVIDED ARTERIAL ROADS IN MELBOURNE

MODELLING NETWORK FLOWStypes of 4-lane road, ditferentiated by spatial location into 'Inner area','Middle suburb' and 'Outer suburb' types after studying the effects oflane use distribution, population density and signalised intersectiondensity in the Beard-McLean model, Table 1 shows the estimated parametersfor the three road types, together with the density of signalisedintersections (in 1979) for the three types. The relationships areshown in Figure 4, The parameters were found using the method given byTaylor (1977)"

PARAMETER

~-----,------------------'-,

c (mi n/km)oJ

5 (veh/h/lane)Signal isedintersectiondensity(i ntn .jkm) (a)

3000

VEH/H

79) TheoreticalData Points

(7)

peed-volume

OLUME

use in OPERAby Kent (1geO),

the ~le1bour neeen tr affi c fl

In thethe data and4) were used

--~

! expressionsnission "rates1 expressionsU\" For;) and unit

I unit traveltype and!Stimated for

TAYLOR AND ANDERSON

BoyequTra

~ea

4- 1

Ba treg(1 )

AkcEtwoConi

Aus'SAE,

REFE

And'modEnumt

AUS1SAE,and

And'Di v,pref

requhi erscheeffedemaandhierThepoteby psyst

moderesplinteby T'or i 9useresUhaslequcent

LARGE SCALE

>En"""FlY' ,,,.,,,.

OPERA to LAIM - Sub-area Trip Demand Estimation

An important linkage within the network flow model hierarchy isthat of providing trip demand (origin-destination) data for a localarea network (LATM) from a larger-scale network (OPERA) .. These data maybe considered as similar to those from a cordon line survey, with thecordon completely surrounding the local area (see Figure 5). The majorcomplication is that the demand data must indicate the connections oftrips entering and leaving the local area.. In the field, a 'number­plate' survey might be used to obtain th,S information .. Within themodelling system other methods may be used, involving the estimation ofan origin-destination matrix from observed (modelled) 1ink flows. Thesemethods are particularly useful for it is difficult to justify the useof trip distribution models for this application, which involves partialnetworks and segments of trips. Various solutions to this intriguingproblem have been proposed recently, and those due to Van Zuylen andWillumsen (1980) and Le Blanc and Farhangian (I982) are perhaps the bestavailable examples. Van Zuylen and Willumsen used inforrr,ation theory toestimate the demand matrix, while Le Blanc and Farhangian explored analternative method which used the flows and costs output from anequilibrium assignment to establish a non-linear mathematical programmingmodel for the sol ution ..

Figure 5. Local Area Network Inside Large-Scale Network

lhe information theory approach offers some advantages in that:la) it can use partial data (Le. it does not require knowledge of

all link flows) to provide estimates, although the estimationprocedure is impr'Qved if more link flows are available;

(D) it is a general procedure which can use either modelled orobserved flows as its inputs; and

(c) it is known to produce unbiased estimates making full use of theavailable input data.

Currently the Van Luylen-Willumsen procedure is being implemented toprovide this linkage between OPERA and LATM

146

CONCLUSIONS

Anderson, M.. (1981). Validation of an arterial road traff1c slmulationmodel. Monash University, Department of Civil Engineering working papernumber 13/HL

Beard, CL. and Mclean, J.R, (1974) .. Speed flow relat10nship for nine4-lane roads in Melbourne Proc. 7th ARRB ConL 7 (4), pp 39-53.

Boyce, D.E , Janson, B,.N,., and Eash, R"W .. (198I) , The effect ofequilibrium trip assignment for different link congestion functionsTransportation Research I5A, pp 223-232

147

and Watson, H.C (1982) ReI ation betweenProgram and Papers.. 2nd Joint SAE/ARRB

Emissions, Melbourne, May, Paper No 7

REFERENCES

Akcelik, R., Richardson; AJtwo fuel consumption model s ..Conf on Traffic, Energy and

MOOELLING NETWORK FLOWSlOteractions Between OPERA and TOPAZ82

The objective of combining assignment and land use/distributionmodels is to seek a network equilibirum in which travel dema~d varies inresponse to travel conditions on the network .. The elements of theinterchange are the travel demand origin-destination matrices producedby TOPAZ82, which are used by OPERA to simulate network travel, and theorigin-destination trip cost matrix which can be produced by OPERA foruse by TOPAZ82. This work is continuing at present, and preliminaryresults should be available shortly. The feasibility of thiS interactionhas been demonstrated by Luk and Nairn (1980), who are able to locatelequilibrium l regions in which flows and costs varied up to 10 percent about the optimum point.

Anderson, M. (1983). Computer user manual for pro9ram PDLOIF.Divisional Report, C,IRO Division of Building Research, Melbourne (inpreparation) ..

Austral ian Road Research Board (I9BO), Program and Papers .. JointSAE/ARRB Seminar 'Can Traffic Management reduce Vehicle Fuel Consumpt10nand Emissions and affect Vehicle Oes1gn Requirements?', Melbourne, JUly.

Australian Road Research Board (1982)" Program and Papers., 2nd JointSAE/ARRB Conf, on Traffic, Energy and Emissions, Melbourne, May.,

Batten, O.F, (1982) The interregional linkages between natlOnal andregi ona1 i nput-ou tput mode1s. l!:'ter nati ona I Reg i ona I Sc i ence Revi ew 7(I), pp.o3-68 ..

This paper has identified a series of transport problemsrequiring solution using network flow models, and has identified ahierarchy of model levels to consider these problems .. The hierarchicalscheme offers a practical, integrated methodology for studying theeffects and influences of changes to network characteristics, traveldemands and operating rules at a variety of levels, and for the studyand isolation of secondary interactions. Particular models within thehierarchy were described, and communication links between them exploredThe integration of models from various levels was shown to offerpotential for improved land use/transport/environment impact analysis,by providing mechanisms for connecting micra-level models to generalsystem-wide effects.

I emented to

full use of

etwork

I hierarchy isJr a localThese data may~y, with the') .. The majorlnections ofa I numbeY'­

;ithin theestimation of

< flows Thesestify the usenvolves partial5 intriguingZuylen and

"haps the bestltion theory toexplored an

fyom ani ca1 o",oramlni

tages in that:knowledge of

e estimationI able;delled or

148

Davidson, K.B. (1966) .. A flow travel time relationship for use intransportation planning, Proc. 3rd ARRB Conf.. 3(1), pp 1~3-194"

Taylemodel

Vi neto TRese

Tay1,Di v i ~

Vanmatrpp .. 2:

Tay1,tran'Traf·

Tay1,use

Tayl'Di vi ~

stephModelinter

Tayl eDivi~

Sharpfor pResea

TaylcparanScier

Sandsspati~

Seddodiffu

Nguye~

fie1waMe1wa,

Luk, ,distrAdela

(1981) The impact of electric vehicles on petrolProc. lost. Engrs. Aust. Transportation Conf,", Br';sbane,

Gipps, P.GconsumptionSeptember "

TAYLOR AND ANDERSONBrotchie J .. F, Dickey, J,W. and Sharpe, R, (1980), TOPAZ: GeneralP1anning'Technique and its Applicatior", (Springer-Verla9: New York),

Ceder, A. (1979). An algorithm to assign pedestrian groups disperSingat public gatherings based on pedestrian-traffic mOdelling 6£21 ledMathematical Modelllng 3, pp .. 11b-124,

Gipps, P.. G (1982), Computer user manual from program MULTSI~I,Dlvisiona1 Report, CSIRO Division of Building Research, Melbourne,

Environment Protection Authority of Victoria (1979) , Draft stateenvironmental protection policy foy' the air environment of Victoria andexplanatory document" EPA, 11e1bourne,

Fruin, J J, (1971). Designing for pedestrians: a 1evel-of-serviceconcept Highway Research Record 355, pp"1-15,,

Gipps, P,G, and Wilson, B.G. (1980)" MULTSIM: a computer package forsimulating mUlti-lane traffic flows. Proc. 4th Biennial Conf..,Simulation Society of Australia, Brisbane, August..

Hall, M.D,., Van Vl iet, 0", and >li Ilumsen, L.G" (1980). SATURN: asimulation-assignment model for the evaluation of traffic managementschemes Traffic Engineering and Control 21, pp, 168-177

Herman, R, and Prigogine, 1. (1979) .. A two-fluid approach to towntraffic, Science 204, pp,,148-151.

Hoban, C,J and McLean, J.R. (1982). Progress with rural trafficsimulation, Proc. 11th ARRB Conf, 11(4), pp,23-33

Kent, J,H (1980), Relationships between driving cycles and urbandrivin9 patterns Program and Papers, Joint SAE/ARRB Seminar 'CanTraffic Management reduce Vehicle Fuel/Consumption and Emissions andaffect Vehicle Design Requirements?', Melbourne, July,

Kent, JH, Post, K, and Tomlin, J,A .. (1982) .. Fuel I consumption andemission modelling in traffic links Program and Papers .. 2nd JointSAE/ARRB Conference on Traffic, Energy and Emissions, Paper no ID

Lawlor, L.C. and Brown, ItA. (1980)" Consumption of transport ener9Y inAus tra 1i a 1975-/6 Bureau of Tr ansport Economi cs Dccasi ona I Paper 37, AGPS

Le Blanc, L.J" and Farhangian, K, (1982)" Selection of a trip tablewhich reproduces observed link flows Transportation Research 16B (2),pp,83-88,

Leonard, U.. R" and Tough, J.B (1979). Validation work on CONTRAM _ amodel for use in the design of traffic management schemes, Proc. PTRCSummer Annual Meeting, University Of >larwick, July

Luk, J,Y"K., Markarov, A and >ligan, M.R,. (1979)" ARRBTRAFIC: programuser manual - equilibrium traffic assignment for fixed demands"Technical Manual ATM No 4A, ARRB

149

MODEl.LING NETWORK FLOWS

Interaction between land-useInst. Engrs. Aust. Engineering Conf ,

Luk, J"Y.K and Nairn, R.J. (1980) ..distribution and assignment.. Proc.Adelaide, April, pp .. 195-201..

Melway (1979). ,treet directory of Greater ~lelbourne Edition 13.Melway Publ ishing Pty .. Ltd., Melbourne.

Nguyen, S. (1974)" An algorithm for the traffic assignment problem.Transportation Science 8, pp 203-216"

,ands, P. and Ciolek, T..M. (1979). Computer simulation and the study ofspatial behaviour of pedestrians: an exploratory exampleMan-Environment Systems 9(6), pp,,281-292 ..

Seddon, P.A. (1972). Another look at platoon dispersion - 3: Thediffusion theory. Traffic Engineering and Control 13 (9), pp .. 338-39U ..

Sharpe, R.. , Wilson, B.. 8. and Pallot, R.G" (1982). Computer user manualfor program TOPAZ82. Divisional Report, CSIRO Division of Buil dingResearch, Melbourne"

,tephanopoulos, G., Michalopoulos, P.. G" and Stephanopoulos, G (1979) ,Modelling and analysis of traffic queue dynamics at signalizedinter sections, Transportation Resear ch 13A, pp .. 295-307 ..

Taylor, ~1.A.P. (1977).. Parameter estimation and sensitivity ofparameter values in a flow-rate/travel-time relation. TransportationScience 11 (3), pp .. 251-256

Taylor, ~l.AP. (1979)" Evaluating the performance of a simulationmodel.. Transpmrtation Research 13A (3), pp .. 159-173 ..

Tay Ior, M.A. P. (1982a). Computer user manual for program LATltDivisional Report, GSIRO Division of 8uilding Research, ~lelbourne.

Taylor, M.A.P. (1982b). Computer user manual for program OPERADivisional Report, CSIRO Division of 8uilding Research, Melbourne"

Taylor, ~1.A.P. (1982c). Computer user manual for program URPOL.Divisional Report, CSIRO Division of Building Research, Melbourne

Taylor, ~I..A.P. and Anderson, M. (1982)" Modelling pollution and energyuse in urban road networks. Proc. 11th ARRB ConL 11 (6), 1-17.

Taylor, M.A.P. and Gipps, P.G (1982)" On the mOdelling of flows intransport systems.. Program and Papers .. Joint SAE/ARRB ConL onTraffic, Energy and Emissions, Melbourne, May, Paper NO .. 23.

Van Zuylen, H.J. and Willumsen, LG. (198U). The most likely tripmatrix estimated from traffic counts" Transportation Research 14B (3),pp .. 281-293.

Vincent, R.,A,., ~litchell, A.!.. and Robertson, D.. !.. (198U) user's guideto TRANSYT/~. TRRL Laboratory Report l.R888, Transport and RoadResearch l.aboratory, Growthorne, Berkshire

]spor t energylOa1 Paper 37,

and urban~rni na r I Cannissions and

option and2nd Joint

Jer no ,10"

-of-service

JLTSHlqelbourne"

ups dispersingng .. Applied

ch to town

,etr 0 1'If '" Br i sbane,

I trip table;earch 16B (2),

" package for-f.onf ., ,

~Z: General,g : New York)"

(AFIC:~mands "

1 CONTRAM - a; . Proc. PTRG

I traffic

for use in1~3-194

1ft state'f Victoria and

ITURN: a:: management