Professional Documents
Culture Documents
AND
COMPUTER
MODELLING
PERGAMON Mathematical and Computer Modelling 35 (2002) 643-656
www.elsevier.com/locate/mcm
1. I N T R O D U C T I O N
The classical network models of transportation systems are based on the assumptions of sta-
tionarity. This assumption, which is acceptable for many applications (i.e., city planning), does
not allow us to simulate satisfactorily certain types of transportation systems such as heavily
congested urban road networks. Phenomena such as the formation and dispersion of queues are
relevant and cannot be reproduced by traditional static models (see [4-6] for a good introduction
to the problem). Traffic engineer scientists have been studying for years the modelling of the
dynamic behaviour of many of the components of a transportation system, but only recently the
research efforts in traffic modeling and traffic assignment have been focused on the modelling of
traffic dynamic behaviour in a complete network system.
A large number of models and procedures have been proposed that are usually referred to, in the
literature, as (within-day) dynamic traffic assignment (DTA) models. These models can be used
both to evaluate traffic flows and, what is more relevant, to simulate the effects of regulation
strategies on users' behaviour. These models originate from static network traffic assignment
models based on stochastic or deterministic user equilibrium. In many early cases, dynamic
traffic assignment models were merely an extension of the concepts contained in static models.
But the extension of within-day static models to take into account within-day dynamics is by
no means straightforward, since within-day dynamic supply modeling requires completely new
definitions and formulation of the problem [7]. In order to perform DTA, in fact, it is necessary
0895-7177/02/$ - see front matter (~) 2002 Elsevier Science Ltd. All rights reserved. Typeset by A.h/cS-TEX
PII: S0895-7177(01)00187-X
644 V. ASTARITA
to solve the dynamic network loading problem (DNL) that has been indicated (by the specialists
of static traffic assignment) as the reproduction of within-day variable link performances given a
corresponding O/D demand and users'choice model.
The dynamic network loading problem, in other words, is the reproduction of the traffic flow
motion on the network. It has to be studied with time advancing mathematical models that are
commonly referred (by traffic engineering specialists) as traffic simulation models.
The DNL problem, so far, has been studied with a great number of different approaches that
are sometimes indicated as:
• microsimulation models [8-11];
• mesosimulation models: heuristic generalization of within-day static methods [12,13]; exit-
function methods [14-20]; packet-approaches and simulation procedures [2,21-31]; contin-
uous time link models [32-36], where often the problem of finding conditions for the respect
of FIFO on a link is discussed;
• macrosimulation models (continuous in time and space flow models [37-42]).
The above-mentioned types of DNL are not so dissimilar from each other: there is a continuous
spread among the conceivable network franc simulation models. Due to a different discretization
of time, space, and demand, sometimes different simulation procedures are not recognized as being
originated by the same underlying model (very interesting the case in [39,41,42]). Sometimes the
confusion arises also because a distinct discretization of the same model can lead to quite different
results.
Moreover, two important problems are still open in traffic network simulation:
• a bad representation of traffic dynamics: some of the proposed models do not even address
the significant problem of the backward propagation of congestion;
• the difficulty of collecting experimental data to test (and sometimes even to run) the
models.
Both problems are far from being solved, and many researchers are concentrating their efforts on
this topic (DNL) that has so many practical applications.
The model proposed in this paper has some advantages and also many disadvantages compared
to other approaches. The model is discussed in the following not because it is better than other
approaches, but because it is used as a starting point to clarify that some common problems,
the backward propagation of congestion at the intersection level, can be solved by using
packets to represent traffic vehicles. In fact, the model is a time-advancing algorithm that follows
explicitly the motion of packets on the network, but can be seen also as a discrete solution of
a time advancing continuous analytical differential equation system. The model is based on the
analytical DNL model presented in [3], which is able to deal with the spill-back of congestion and
is based on the preceding link model formulation presented in [19,32,33,35,36].
This paper is organized as follows: in Section 2, a classification of dynamic network loading
(DNL) procedures (based on time, space, and demand discretization) is introduced. Many prob-
lems that arise in DNL models are discussed in Section 3. In Section 4, some general definitions
are introduced. In Section 5, the proposed model is briefly discussed, and at last in Section 6
some analytical properties of the point packet approach are discussed and results are extrapolated
from the proposed procedure (called MICE [2]) to other analytical models.
requirements are respected by a group of similar models, since very similar models may not
respect the same requirements.
The following main modelling approaches can be identified:
(1) microsimulation models,
(2) continuous in time link models,
(3) discrete in time link models,
(4) models following a packet approach, and
(5) macrosimulation models (continuous in time and space models).
According to time, space, and demand discretization, it is possible to represent models in a
three-dimensional space where the x, y, and z axes are time, space, and demand. The 0 value
would be representative of a continuous model, and a finite value would represent a finite step for
the representation of time, space, or demand. A continuous in-time model will be located on the
space-demand plane (y-z). In Figure 1, examples of the above-mentioned models are located by
four points and a plane. Microsimulation models are based on the simulation of the movement
of single users. Microscopic simulation is indicated in reproducing some specific local traffic
situations, such as intersection and parking facilities, or for the evaluation of control strategies
which act on single individual driver behaviour. Some mixed models have been proposed, based
on a microscopic or quasi-microscopic simulations, obtaining macroscopic link characteristics such
as speed and density [28].
Demand
~ n u o u s in t i m e link m o d e l s (2)
Many authors (Friesz et al. [18], Wie et al. [20], Boyce et al. [151, Fernandez and De Cea [431,
and Adamo et al. [3]) have proposed formulations of continuous in time link models that are
based on a space discretization which divides up the path of the users into links which together
form the network. Some of these continuous models have been numerically solved with a time
discretization and so can be related to concepts originated in the discrete in time link model
proposed by Merchant and Nemhauser [14]. Discrete in time link models are based on the two
following equations: x, u, and w are defined in the following, respectively, as number of vehicles
646 v. ASTARITA
on a link, entering and exiting flows; i and j are, respectively, time interval and link number:
Demand
~
_Lightill and Whitham model
nsion dx
/J "-.,
"--Time 1
Cell transmission model Space
Figure 2. The Daganzo cell transmission model [5] originates from the Lighthilland
Whitham's model.
The following two simple categories can be defined for DNL methods and are quite different
in demand (traffic flow) discretization.
• Time-advancing differential equation systems that represent the motion of traffic as being
the motion of a fluid. These equations have the same numerical problems as nonstationary
fluid dynamics equations. In these models, the traffic being a continuous fluid, the FIFO
Network Traffic Flow Simulation 647
rule has to be respected, but in practice when the models are discretized, for solution,
it is very difficult to maintain a complete respect of FIFO rule. These models can be
represented as being on the time-space plane as in Figure 3.
* Time-advancing algorithms that follow explicitly the motion of vehicles on the network.
Sometimes vehicles are grouped into packets. These models can be represented, in Fig-
ure 3, as all the points over the time-space plane.
Demand
Planeof all the modelsbaSe:u:~on
~'Tme . . . . .- Spa2
The main problem with DNL approaches is the bad representation of traffic dynamics, which is
caused by the difficulties in reproducing traffic flow. Traffic flow analysis is complicated both by
the human factor present in drivers' behaviour and because it has some peculiar aspects unusual
to other fluids (some are discussed in the following).
Moreover, the complication of collecting experimental data (even only aggregated), to test the
models, is the cause for so many unrealistic approaches not being rejected.
Traffic flow models are different in dealing with some traffic flow aspects. In this section,
the following problems are discussed and some recommendations are given for good modelling
techniques:
• the representation of bottlenecks and connected flow capacities and storage capacities,
• traffic waves, and
• intersection modelling.
The spill-back of congestion on a network is the propagation of congestion backwards from a link
to its upstream links. It occurs whenever a bottleneck causes a downstream queue to be so large
that it impedes the incoming speed and flow of vehicles. This very common situation in many
urban networks is usually caused by recurrent congestion on a day-to-day basis and is not always
well represented by DNL models• In all analytical approaches to DNL, road bottlenecks can be
reproduced at least at the end of links where a downstream capacity is imposed to flows exiting
from each link. But a great number of microsimulation models do not have explicit capacity
constraints. The idea is that the mechanism that reproduces the movement of vehicles through
lane-changing and car-following rules will give as a result an intrinsic capacity constraint. Road
capacities can be easily measured on the field or inferred from analogue situations, but in most
microsimulation models, capacities are the outcome of not less than ten different (on average)
parameters. Some of these parameters are related to the behaviour of different types of drivers,
648 V. ASTARITA
some are related to the local characteristics of the road, but usually there is no clear relation
with the obtained final implicit capacity value or with the factors identified by the HCM as being
related with capacity. This is perhaps the biggest pitfall of most microsimulation packages and
is based on the assumption that the performance of the optimized parameters set obtained with
calibration provides an approximation of the potential performance of this set in the future. The
calibration of many of these models can be performed only by research institutes and so, as a
result, most of the microsimulation models that are used in common practice are similar to many
calibrated stock trading systems: they work only with data of the past!
Among the others, one remarkable exception is the work of Van Aerde with INTEGRATION
where explicit capacities can be imposed in a microsimulation model [11].
But also the representation of a network with only downstream capacities for the links is too
approximate and may not be representative of real situations. In many DNL models, there has
never been an attempt to give a limit to the flows entering the links through an inflow capacity
and a storage capacity. The resulting flow obtained with these models may exhibit links filled
with enormous numbers of vehicles, indicating that these models are essentially unqualified for
practical application.
Relying on common physical experience, the flow entering a link is not only limited by the
upstream capacity, but when the link is full (all the space is occupied by vehicles waiting to
exit), the entering flow should be smaller than or equal to the exiting flow. And this should be
reflected in a dynamic network assignment model. Then two simple concepts must be taken into
consideration by DNL models:
• the spill-back of queues is caused by limited upstream capacity and/or link storage space;
• the surplus of flow that cannot be received by a saturated link is accumulated on the
preceding links.
So concluding, to well represent traffic, all network models should have explicitly associated
with every link a at least:
• upstream capacities, Cian,
• downstream capacities, C °ut, and
• link storage space, C~s.
The conditions of traffic in one point of the road are originated by the conditions at preceding
instants of time in points situated downstream and upstream. The propagation of a traffic flow
condition is usually called traffic wave. The representation of traffic waves in network traffic
models is necessary. An example: the spill-back of congestion on a network is in fact the linear
back-propagation of one or more congestion waves. Only analytical continuous models are able
to deal well with traffic waves. But the numerical solution of many of these models faces many
practical difficulties that arise from the numerical discontinuities of traffic flow parameters on the
two sides of the waves. Some micro and mesosimulation models, but not many, are also able to
propagate waves in both directions as a consequence of microrules for car-following.
Correct traffic simulation procedures models should be able to propagate waves in both direc-
tions as explained also by Daganzo [6].
Traffic intersections have to be represented allowing congestion and queues to propagate back-
wards. An attempt to identify some of the problems related to intersections and queue propa-
gation can be found in [4]. The problem is to distribute the limited resource of flow capacity
on the downstream links of a node to its upstream links. The intersection representation also
needs to take into account the gap acceptance phenomenon. Microsimulation models are good in
achieving this target. In fact, an analytical representation of intersections that is really complete
has still to be proposed.
In [48], the following sentence is stated: "to be useful, analytic modellers must incorporate all
meaningful physical and behavioural considerations. To do less, would be to yield the problem
of traffic dynamics to simulation specialists, leaving no plausible mathematical constructs to use
Network Traffic Flow Simulation 649
4. D E F I N I T I O N S F O R T H E G E N E R A L
DYNAMIC NETWORK LOADING PROBLEM
T h e dynamic network loading problem can be generally stated mathematically as follows. On
a t r a n s p o r t a t i o n network f~ = (N, A) composed of a set of nodes i, i E N and a set of directed
arcs (links) a, a ¢ A. Traffic originates at nodes o, o E O, o C N and is destined to nodes d,
dED, DcN. We have:
• origin-destination (o-d) pairs designated as: r, r E R C 0 x D;
• a set of paths k, k E KT, t h a t connects o-d pairs r;
• DT(t) as the time varying demand flow rate t h a t uses these paths, traffic departs from
origins in the interval [0, T] and all traffic arrives at destination within the interval [0, T'],
(T' > T);
• the sets of arcs A + and A~- which are, respectively, the forward and backward stars of
node i.
The rules t h a t determine the flow of traffic on the network are different in each traffic model, but
some general variables, t h a t are related to the traffic motion, can be defined in all the proposed
models and are described in this section.
Traffic flow introduced in the models t h a t apply a ttuidodynamic analogy in the representation
of traffic is defined as the flow q(t) through a point section S of the road,
N. of veh. passed through S during [t, t + At]
qs(t) = lim
z~t--0 At
(3)
This definition can be held also in models that follow explicitly the motion of vehicles on the
network as microsimulation and mesosimulation models. From that definition, it is useful to
indicate the flow at the first (entrance) and last section (exit) of every link a (time varying inflow
rates and outflow rates) for o-d pair r, as: Ura(t) and w~(t), respectively, at time t. The number
of vehicles for o-d pair r on link a at time t can be indicated as
x~(t). (4)
We have t h a t the total inflow rates, outflow rates, and vehicles at time t on link a can be defined
as
..(t) = (5)
r
wo(t) = (6)
r
650 V . ASTARITA
and
respectively.
ua(t), wa(t), and Xa(t) are functions of time and are related by the link flow conservation
equation
dxa(t) = ua(t) - wa(t). (8)
dt
Ura(t), Wra(t), and Xra(t) also are related by a link o-d pair flow conservation equation
xa(t) =
f (ua(t) - wa(t)) dt, (10)
• X(t) =
f (u~(t) - w~(t)) at. (11)
A travel time on link a at time t (the travel time of the traffic particle that enters the link at
time t) denoted as
~o(t) (12)
E dt/At
p(x,t),,~= n dx ' (13)
where ~-~ndt is the sum of the times of the n vehicles to cross the slice of road. The space-mean
speed v is the average vehicle speed weighted according to the time for each vehicle to cross the
slice of road, which is equivalent to the ratio of flow to density
/7, • d x
v- •dt" (14)
n
If equations (3), (13), and (14) are defined, on the same slice of r o a d and time interval, the
fundamental speed-flow-density relationship is obtained,
q = vp, (15)
which can be used to obtain the value of the third parameter given the values of the other two
parameters. It is possible also to identify a local conservation equation that is equivalent to (8)
with continuous space and can be obtained from (8) if the length of the link goes to 0,
Op(x, t) Oq(x, t)
----&--- + o~ = 0. (16)
Network Traffic Flow Simulation 651
5. T H E A N A L Y T I C A L F O R M U L A T I O N
OF T H E P R O P O S E D M O D E L
A simple link modelling approach has been developed in [33]. It is essentially the same as that
presented in [19] and in [35], and consists (for a single destination) of the following system (17)
of differential equations that constitutes a complete link model:
proposed model
{ @ t t) -- u ( t ) - w(t),
T(t) = T(x(t)),
~(t + r(t)) = ~(t)
1 + dr(t)/dt'
(17)
where each link has the following characteristics at time t: upstream flow u(t), downstream
flow w(t), and a total number of users x(t) on the link, where x(t), u(t), and w(t) are continuous
functions of time and r(t) is the travel time at time t (travel time of the particle that enters the
link at time t). This simple model has been studied to establish the FIFO conditions and the
field of applicability also for u(t) not continuous:
• Lipschitz continuous functions have been investigated in [34] and in [36];
• Dirac functions have been studied in [1].
The FIFO condition can be stated formally as
which is equivalent to
W(t) = U(t + 7(t)), (19)
where U(t) and W(t) are, respectively, the total number of vehicles entered (exited) at time t
w(t) =
/o' w(~) d~. (21)
From the results presented in [1], it is clear that the model can always be applied if FIFO
violations are accepted. In this case, the third equation in (17) (equivalent to FIFO rule) has to
be changed in ¢ *
Links are composed of a running segment and of a queuing segment. In terms of input values and
output values, the link model has the general structure indicated in Figure 4. It establishes partial
supply and demand flow rates at time t* as a function of the inflow patterns at time t < t*. Flows
at each node at time t respect the conservation condition, as in all network models, except that
flows at nodes are constrained by the interactions between demand of entering links and supplies
of exiting links. The node model has the structure shown in Figure 5. It establishes the flows at
time t, once supply and demand flow rates at time t are known. The node problem pertains not
only to flow based dynamic network loading (DNL) models, but to every traffic flow model that
considers the spill-back of queues. The limited resource of flow capacity on downstream links
has to be distributed between users in a proper way. The problem can be expressed as a linear
programming problem (see [3]), where the objective is to maximize the total flow crossing the
node, at each time t, and the constraints are the local flow supplies and demands at the node.
Output
Linkmodel
a.,,(t)
t=t*
~ Node model
,' ~t)=,, .b(t)
The maximization of flow comes from entropy considerations and is related only to the mod-
elling of flow propagation. The control strategies to be applied at intersections (i.e., traffic lights)
can be modelled by changing in time (dynamically) capacities and other characteristics of the
links.
The complete analytical network model has been introduced in [3], and turns out to be a
formidable set of equations that cannot be solved analytically except for some trivial cases. The
analytical model proposed is the simplest that can represent at a link level spill-back propagation:
the maximization of flow at the node is the mechanism that allows waves to propagate in both
directions, but is also the cause for the complicated formulation adopted. In order to compute
flows which respect the constraints, some approximate methods are needed. MICE, introduced
in [2], is the simulation procedure used for the solution of the analytical model [3]. The analytical
model is obtained from the Lighthill and Whitham model by discretizing space into links, and
MICE is obtained from the analytical model when demand is discretized (see Figure 6).
6. T H E S O L U T I O N OF ANALYTICAL
MODELS WITH DEMAND DISCRETIZATION
In [2], a new dynamic network loading model has been presented (called MICE). In MICE, users
belonging to the same O/D pair are supposed divided into subsets by the common leaving interval
Network Traffic Flow Simulation 653
Demand
~ k length
Figure 6. The analytical link model proposed in [3], and the discretized solution
MICE.
and followed path; these subsets are called packets. In particular, in the proposed model, a packet
is assumed to be "physically" a point on the network. Packets on the network are represented
with points, and the link travel time, for each packet, is that established at the moment the
packet reaches the beginning of the link. Travel time depends on the number of users on the
link at that time plus the packet itself. This last point is very important, because otherwise,
in spite of the fact that from an analytical point of view, the travel time function corresponds
to a bottleneck, packets may be able to exit from the link in such a way that capacity will not
be respected. Capacity in the model is in fact automatically respected (mean values over time)
without the need of additional constraints.
In [1], problem (17) is studied in the case of u ( t ) not continuous and reducible to impulse
functions (Dirac functions). The Dirac function 6~ is an impulse function with intensity equal
to i (see Figure 7). It is called the Dirac delta function after the Nobel physicist Dirac. Intuitively,
it is possible to think of 6~ being zero except at t = a, where 6~ = oc, but, still, the following
integral has value i:
+ _ ~ 6ia(t) dt = i. (24)
+co
(~(t) H ai (t)
> >
a
t a t
The 5~ is derived from the Heaviside function H~(t) (step function) defined by (see Figure 8)
7. C O N C L U S I O N
In this paper, some of the problems of dynamic network loading procedures have been discussed.
Some considerations on MICE results have been also discussed using the results of Chabini and
Kachani [1,34] on the case of u(t) not continuous and reducible to Dirac functions.
Network Traffic Flow Simulation 655
Demand
,t
Micro or packet solution to Lightill and Whitham model
~and Whithmodel
am
REFERENCES
1. I. Chabini and S. Kachani, Analytical dynamic network loading I: Dirac modeling, Optimization Days,
Montreal, May 1999.
2. V. Adamo, V. Astaxita, G.E. Cantarella and E. Cascetta, A doubly dynamic traffic assignment model for
planning application, 14 th International Symposium on Theory of Traffc Flow, (1999).
3. V. Adamo, V. Astarita, M. Florian, M. Mahut and J.H. Wu, Modelling the spillback of congestion in link
based dynamic network loading models: A simulation model with application, 14 th International Symposium
on Theory of Traffic Flow, (1999).
4. E. Cascetta, Metodi Quantitativi per la Pianifieazione dei Sistemi di Trasporto, CEDAM, Italy, (1990).
5. C.F. Daganzo, The cell transmission model, Part I and Part II, Traspn. Res. 29B (2), (1995).
6. C.F. Daganzo, Properties of link travel time functions under dynamics loads, Traspn. Res. 29B, (1995).
7. E. Cascetta and G.E. Cantarella, A day-to-day and within-day dynamic stochastic assignment model, Traspn.
Res. 25A (5), 277-291, (1991).
8. J. Barcelo, J.L. Ferrer and L. Montero, AIMSUN: Advanced Interactive Microscopic Simulator for Urban
Networks. Vol. I: System Description, and Vol. II: Users Manual, Departamento de Estadistica e Investiga-
cion Operativa, Facultad de Informatica, Universidad Politecnica de Cataluna, (1989).
9. J. Barcelo, The parallelization of AIMSUN2 microscopic traffic simulator for ITS applications, Presented at
the 3 rd World Conference on Intelligent Transport Systems, (1996).
10. L. Rilett, C. Benedek, H. Rakha and M. VanAerde, Evaluation of IVHS options using CONTRAM and
INTEGRATION, In First World Congress on Applications Transport Telematics and Intelligent Vehicle
Highway Systems, Paris, France, (1994).
11. M. VanAerde, S. Yagar, A. Ugge and E.R. Case, A review of candidate freeway arterial corridor traffic
models, Transportation Research Record 1132, TRB, (1988).
12. R. Hammerslag, A three-dimensional assignment in the time-space, In UTSG Annual Conference, London,
(1988).
13. B.N. Janson, Dynamic traffic assignment for urban road networks, Traspn. Res. 25B (2/3), 143-161, (1989).
14. D.K. Merchant and G.L. Nemhauser, A model and an algorithm for the dynamic traffic assignment problems,
Transportation Science 12 (3), 183-207, (1978).
15. D.E. Boyce, B. Ran and L.J. LeBlanc, Dynamic user-optimal traffic assignment model: A new model and
solution technique, In First 7~ennal Symposium on Transportation Analysis, Montreal, Canada, June 6-11,
1991.
16. M. Carey, A constraint qualification for a dynamic traffic assignment model, Operations Research 35 (5),
55-58, (1986).
17. M. Carey, Optimal time-varying flows on congested networks, Operations Research 35 (5), 58 69, (1986).
656 V. ASTARITA
18. T.L. Friesz, J. Luque, R.L. Tobin and B.W. Wie, Dynamic network traffic assignment considered as continuous
time optimal control problem, Operations Research 37, 893-901, (1989).
19. T.L. Friesz, D. Bernstein, T.E. Smith, R.L. Tobin and B.W. Wie, A variational inequality formulation of the
dynamic network user equilibrium problem, Operations Research 41, 179-191, (1993).
20. B.W. Wie, T.L. Friesz and R.L. Tobin, Dynamic user optimal traffic assignment on congested multi destina-
tion networks, Traspn. Res. 24B (6), 431-442, (1990).
21. V. Adamo and V. Astarita, Un nuovo modello di caricamento dinamico del traffico con gestione dello spill-back
e degli incidenti, Rivista del C.S.S. 7". Luglio-Settembre, (1996).
22. V. Adamo, V. Astarita and M. di Gangi, A dynamic network loading model for simulations of queue effects and
while-trip re-routing, In 24 th European Transport Forum PTRC, September 2-6, 1996, Brunel University,
Uxbridge, (1996).
23. E. de Romph, H.J.M. van Grol and R. Hamerslag, A dynamic traffic assignment model for short-term predic-
tions, In Seminar on Urban Traffic Networks, Capri, July 1992.
24. M. di Gangi, Continuous flow approach in dynamic network loading, In Second International Capri Seminar
on Urban Traffic Networks, Capri, July 5-8, 1992.
25. M. di Gangi and V. Astarita, Structure of a dynamic loading model for the evaluation of control strategies,
In T R I S T A N H Second Triennial International Symposium on Transportation Analysis, Capri, June 23-28,
1994.
26. M. di Gangi, V. Adamo and V. Astarita, Dynamic network loading model and algorithms for explicit simu-
lation of queue-effects and while-trip re-routing, In 7 th WCTR World Conference on Transport Research,
Sydney, July 1995, (1995).
27. D.R. Leonard, J.B. Tough and P.C. Baguley, CONTRAM: A traffic assignment model for predicting flows
and queues during peak periods, TRRL Report 841, Crowthorne, (1978).
28. H. Mahmassani and P. Chen, Dynamic interactive simulator for the study of commuter behavior under
real-time traffic information supply strategies, Traspn. Res. Record 1413, 12-21, (1993).
29. M.J. Smith and M.B. Wisten, Parallel dynamic traffic equilibrium assignment, Traffic Engineering and
Control, (December 1993).
30. M.J. Smith and M.B. Wisten, A distributed algorithm for the dynamic traffic equilibrium assignment problem,
In 13 th International Symposium on Theory of Traffic Flow, Lyon, July 1996, pp. 385-408, Elsevier, (1996).
31. S. Yagar, CORQ--A model for predicting flows and queues in a road corridor, In Transportation Research
Record, Volume 533, TRB, pp. 77-87, National Research Council, Washington, DC, (1975).
32. V. Astarita, Flow propagation description in dynamic network loading models, In Proceedings of IV Interna-
tional Conference on Application of Advanced Technologies in Transportation Engineering (AATT), Capri,
pp. 599-603, American Society of Civil Engineers, (1995).
33. V. Astarita, A continuous time link model for dynamic network loading based on travel time function, In
13 th International Symposium on Transportation and Traffic Theory (13 th ISTTT), Lyon, July 1996.
34. I. Chabini and S. Kachani, Analytical dynamic network loading II: Lebesgue's integrable bounded flow rates,
In Optimization Days, Montreal, May 1999.
35. J.H. Wu, Y. Chen and M. Florian, The continuous dynamic network loading problem: A mathematical
formulation and solution method, Traspn. Res. 32B, 173-187, (1998).
36. Y. Xu, J.H. Wu, M. Florian, P. Marcotte and D.L. Zhu, New advances in the continuous dynamic network
loading problem, Publication CRT-96-26, (1996); Transportation Science (to appear).
37. J.P. Lebacque, The Godunov scheme and what it means for first order traffic flow models, In 13 th Interna-
tional Symposium on Theory of Traffic Flow, pp. 647-677, Lyon, July 1996, Elsevier.
38. M. Lighthill and G. Whitham, On kinematic waves I and II: A theory of traffic flow on long crowded roads,
Proc. Royal Society, London, Series A 229, 317-345, (1955).
39. G.F. Newell, Comments on traffic dynamics, Transportation Research, (1989).
40. M. Papageorgiou, Dynamic modelling, assignment, and route guidance in traffic networks, Traspn. Res. 24B
(6), 471-495, (1990).
41. P. Ross, Traffic dynamics, Traspn. Res. 22B (4), 234-252, (1988).
42. P. Ross, Response to Newell, Traspn. Res. 23B (5), 123-145, (1989).
43. J.E. Fernandez and J. de Cea, Flow propagation description in dynamic network assignment models, In
T R I S T A N H Triennal International Symposium on Transportation Analysis, Capri, June 1994.
44. E. Codina and J. Barcel, A system optimal dynamic traffic assignment model with distributed parameters,
In T R I S T A N H Triennal International Symposium on Transportation Analysis, Capri, June 1994.
45. H. Payne, Models of freeway traffic and control, In Simulation Councils Proc., Simulation Councils, La Jolla,
CA, (1971).
46. B.D. Greenshields, A study in highway capacity, Highway Research Board, Proceedings, (1935).
47. H. Payne, FREEFLO: A macroscopic simulation model of freeway traffic, TRR 722, (1979).
48. T.L. Friesz, S.V. Lall and R.R. Stough, Traffic Network Dynamics: Alternative Mathematical Formulations,
TRB, National Research Council, Washington, DC, (2000).