From Traffic and Pedestrian Follow-The-Leader Mode
From Traffic and Pedestrian Follow-The-Leader Mode
From Traffic and Pedestrian Follow-The-Leader Mode
net/publication/311612645
CITATIONS READS
15 1,048
4 authors:
Some of the authors of this publication are also working on these related projects:
A Machine Learning Approach for Pushing Behavior Detection in Large-Scale Events View project
All content following this page was uploaded by Antoine Tordeux on 11 January 2017.
Abstract
In this work, we derive first order continuum traffic flow models from a microscopic
delayed follow-the-leader model. Those are applicable in the context of vehicular traffic flow
as well as pedestrian traffic flow. The microscopic model is based on an optimal velocity
function and a reaction time parameter. The corresponding macroscopic formulations in
Eulerian or Lagrangian coordinates result in first order convection-diffusion equations. More
precisely, the convection is described by the optimal velocity while the diffusion term depends
on the reaction time. A linear stability analysis for homogeneous solutions of both continuous
and discrete models are provided. The conditions match the ones of the car-following model
for specific values of the space discretization. The behavior of the novel model is illustrated
thanks to numerical simulations. Transitions to collision-free self-sustained stop-and-go
dynamics are obtained if the reaction time is sufficiently large. The results show that the
dynamics of the microscopic model can be well captured by the macroscopic equations. For
non–zero reaction times we observe a scattered fundamental diagram. The scattering width
is compared to real pedestrian and road traffic data.
Keywords: First order traffic flow models, micro/macro connection, hyperbolic conservation
laws, Godunov scheme, numerical simulation.
AMS: 35F20, 70F45, 90B20, 65M12.
1 Introduction
Microscopic and macroscopic approaches for the purpose of vehicular traffic flow modelling have
been often developed separately in the engineering community [43, 22, 5, 25]. Similar models
can also be used in the description of pedestrian dynamics [40, 10, 1]. Typically, microscopic
models are based on the so-called “follow-the-leader” strategy and they are stated as (finite of
infinite) systems of Ordinary Differential Equations (ODEs). They are generally based on speed
or acceleration functions which depend on distance spacing, speed, predecessor’ speed, relative
speed and so on. One of the simplest approach is a speed model solely based on the spacing,
firstly proposed by Pipes [38]
ẋi (t) = W (∆xi (t)), (1)
where ∆xi (t) = xi+1 (t) − xi (t) denotes the spacing between the vehicle (i) to its predecessor
(i + 1) and W (·) stands for the equilibrium (or optimal) speed function depending on the spac-
ing. The microscopic models are discrete in the sense that the vehicles or pedestrians i ∈ Z are
∗
Forschungszentrum Jülich GmbH and Bergische Universität Wuppertal, Germany ([email protected]).
†
INRIA Sophia Antipolis–Méditerranée, France ([email protected]).
‡
Rheinisch–Westflische Technische Hochschule Aachen, Germany ([email protected]).
§
Forschungszentrum Jülich GmbH and Bergische Universität Wuppertal, Germany ([email protected]).
1
individually considered. A macroscopic description consider the flow of vehicles or pedestrians
(in the following also referred to as agents) as a continuum in Eulerian or Lagrangian coor-
dinates. For instance in the most classical Eulerian time-space framework, the main variables
are the density, the flow and the mean speed. The simplest approach is the scalar hyperbolic
equation of the celebrated Lighthill-Whitham-Richards (LWR) model [29, 39]
Here ρ is the density, V (·) is the equilibrium speed function which is assumed to depend only
on the density. The flow f (ρ) = ρV (ρ) is given by the product of the density times the mean
speed. The model is derived from the continuity equation for which the flow is supposed in
equilibrium. The microscopic and macroscopic models Eq. (1) and Eq. (2) well reproduce
shock-wave phenomena for Riemann problems. Yet such models are not able to describe the
observed transition to scattered flow/density relation (the fundamental diagram) with hysteresis
and self-sustained stop-and-go phenomena (see [44, 24, 9] and Fig. 1). This is due to the fact
that spatially homogeneous regime are always in the equilibrium solutions and determined by
the functions W (·) and V (·), respectively.
Figure 1: Empirical fundamental diagrams. Left, [23, Figure 1] and right, [48, Figure 5].
Therefore, the microscopic behavior is modified by introducing reaction and relaxation times.
The simplest following model of this type may be the delayed model by Newell [31]
with τ the reaction time (if positive). Applying a Taylor expansion in the l.h.s. of the delayed
speed model Eq. (3), we obtain the second order ‘optimal velocity model’ (OVM) introduced
by Bando et al. in [4]. The OVM has limit-cycles in stationary states, with self-sustained prop-
agation of non-linear stop-and-go waves, and hysteresis curves in the fundamental flow/density
diagram (see [34, 35]). Macroscopic second-order models comprised of systems of hyperbolic
equations are also able to reproduce non-linear stop-and-go waves and scattering of the funda-
mental diagram. One of the first approach is the one by Payne and Whitham [37, 46]. The
model can be derived from the microscopic Newell model Eq. (3). The main drawback of this
model is that, as pointed out by Daganzo [13], the speed and the density could yield negative
values and are not bounded. Note that this drawback is also observed with follow–the–leader
models like the OVM and is referred as collision between the vehicles (see for instance [14, 36]
or [43, Chap. 15]). Aw and Rascle have corrected this issue by replacing the space derivative
2
of the ‘pressure’ by a convective derivative [3] (AR model). Nowadays extensions of the AR
model such as the ARZ, GARZ or generalized models [21, 47, 6, 17, 41, 16], as well as two phase
models coupled with the LWR model [11, 19, 12, 7], are used to describe transition to congested
traffic with scattered fundamental diagrams and self-sustained non-linear shock waves. A gen-
eral framework is the generic second order model (GSOM) family introduced in [27, 28]. Most
of the approaches are a posteriori based on the continuous description.
In this article, we derive minimalist macroscopic traffic flow models of first order from a micro-
scopic speed model to describe stop-and-go wave phenomena and scattering of the fundamental
diagram. The use of first order models allow us to ensure by construction that the speed and
the density remain positive and bounded. The starting point is a OV microscopic model of
first order including a reaction time parameter. We show in Sec. 2 that the corresponding
macroscopic model results in a convection-diffusion equation. The macroscopic model is dis-
cretized using distinct Godunov and Euler-based schemes and the linear stability conditions for
the homogeneous solutions of these numerical schemes are provided in Sec. 3. The conditions
match the ones of the car-following model for specific values of the spatial discretization step.
Simulations are carried out in Sec. 4. Systems with different initial conditions are numerically
solved. Further, we compare with data of realistic traffic flow as well as pedestrian flow.
3
models is useful to fully understand the dynamics. In [2], Aw et al. established the connection
between a microscopic car-following model and the second-order AR macroscopic traffic flow
model. The rigorous proof, based on a scaling limit where the time and space linearly increase
while the speed and the density remain constant, assumes homogeneous conditions. We use
here the same methodology considering the local density ρi (t) around the vehicle or pedestrian
(i) and at time t > 0, as the inverse of the spacing
1
ρi (t) := . (7)
∆xi (t)
The density could also be normalized by multiplication with `. Here, we prefer to keep the
unit of one over length as density to ease the comparison with the classical models. Then, the
microscopic model reads
1 1 1
ẋi (t) = W −τ W −W =: Ṽ (ρi+1 (t), ρi (t)), (8)
ρi (t) ρi+1 (t) ρi (t)
Hence, we observe that in the rescaled time and in the limit ∆y → 0 the microscopic model is
an upwind discretization of the following macroscopic equation
1 ρ
∂t − ∂y V = 0. (12)
ρ 1 − τ ρ∂y V (ρ)
The upwind or Godunov scheme is the most mathematically reasonable discretization provided
τ is sufficiently small due to the decreasing behavior of V for suitable OVM functions W. Up to
4
second–order in τ we approximate (12) by Taylor expansion and obtain a convection–diffusion
model as
1
∂t − ∂y V (ρ) = τ ∂y (ρV 0 (ρ))2 ∂y ρ . (13)
ρ
The relation between the density in Lagrangian coordinates Rand Eulerian coordinates is given
x
by the coordinate transformation (t, y) → (t, x) where y = −∞ ρ(t, x)dx. Note that y counts
the number of vehicles/pedestrians up to position x in Eulerian coordinates. In the Eulerian
coordinates (t, x), the macroscopic model Eq. (12) reads
ρ
∂t ρ + ∂x ρV = 0. (14)
1 − τ ∂x V (ρ)
The model could be seen as an extension of the LWR model Eq. (2) with a modified speed-
density relationship ρ 7→ V (ρ/(1 − τ ∂x V (ρ))). For illustrating the behavior of this modified
speed-density mapping, we set I := τ ∂x V (ρ)) and we define V : (ρ, I) 7→ V (ρ/(1 − I)). The
fundamental diagrams obtained for a constant term I ∈ {−0.3, 0, 0.3} and for a speed function
V : ρ 7→ max{0, min{2, 1/ρ − 1}} are shown on Figure 2. Note that for constant (in space)
densities ρ (and/or for τ = 0), the additional term I vanishes and we recover the classical LWR
model.
2.0
I=
0.8
0.3
1.5
0
0.6
ρ V(ρ, I)
-0.3
V(ρ, I)
1.0
0.4
0.5
0.2
0.0
0.0
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0
ρ ρ
Figure 2: Illustration for the fundamental diagram V : (ρ, I) 7→ V ρ/(1 − I) obtained in
the macroscopic model (14) with constant inhomogeneity I ∈ {−0.3, 0, 0.3} and V : ρ 7→
max{0, min{2, 1/ρ − 1}}.
5
Z x
where ρ0 ∈ L1 (R)∩BV(R). By defining D(ρ) := −τ (ρV 0 (ρ))2 ∂x ρ dy, we obtain an equation
−∞
similar to the one considered in [8], say
One can verify that D(ρ0 ) is absolutely continuous on R and that ∂x D(ρ0 ) ∈ BV(R). In
(15), the l.h.s. is the LWR model with additional diffusion proportional to the reaction time
parameter τ and that can be either negative or positive. More precisely the diffusion is negative
in deceleration phases where the density get higher upstream, and it is positive in the opposite
acceleration phases. This type of diffusion seems to induce an instability of the homogeneous
(constant) solutions and the formation of oscillations (i.e. jam waves). The diffusion coefficient
(ρV 0 (ρ))2 depends on the density and the fundamental diagram. In fluid dynamics the coefficient
is a characteristic for the flexibility of the random movement responsible for the diffusion. In
traffic flows, comparable diffusion-convection forms have been used in [30, 8]. We refer the
interested reader to [8] (and references therein) for a proof of existence and uniqueness of the
solution to Eq. (15)-(16). In the following we analyze the linear stability of homogeneous
solutions for the macroscopic model.
Proposition 3.1 The homogeneous configurations for which ρ(x, t) = ρe for all x and t are
linearly stable for the continuous traffic model Eq. (15) if and only if
τ < 0. (18)
6
and spatial discretization steps and use the Godunov scheme [20] for the discretization of the
density
dt
ρi (t + dt) = ρi (t) + fi−1 (t) − fi (t) (20)
dx
where fi denotes the flow at cell boundary and has to be determined. For this aim, we introduce
the demand and supply functions from the flow-density fundamental diagram f : ρ 7→ f (ρ) :=
ρV (ρ) as first proposed in [13, 26] and that read respectively
and we define the Godunov flux as G(x, y) := min{∆(x), Σ(y)}. We are now ready to propose
three different strategies to compute the boundary flows fi . The first two methods discretize
the linearized model Eq. (15) using a splitting scheme which treats separately the convection
and the diffusion terms. The last scheme is a simple discretization of the exact macroscopic
model Eq. (14).
1. The Godunov/Euler scheme: a Godunov scheme for the convection term and an
explicit Euler scheme for the diffusive term of the linearised model Eq. (15):
(1) τ
fi = G(ρi , ρi+1 ) + (ρi V 0 (ρi ))2 (ρi+1 − ρi ). (22)
dx
Such a scheme is the one used in [2].
2. The Godunov/Godunov scheme: a Godunov scheme for the convection term and a
Godunov scheme for the diffusion term of the Taylor–expanded model Eq. (15):
(2) τ
fi = G(ρi , ρi+1 ) + ρi V 0 (ρi ) G(ρi+1 , ρi+2 ) − G(ρi , ρi+1 ) . (23)
dx
3. The Godunov scheme: a Godunov scheme for the modified convection term in the
exact macroscopic model Eq. (14):
!
(3) ρi ρi+1
fi = G τ
, τ
. (24)
1 − dx V (ρi+1 ) − V (ρi ) 1 − dx V (ρi+2 ) − V (ρi+1 )
τ
Note that this scheme is valid if 1 − dx V (ρi+1 ) − V (ρi ) > 0 for all ρi and ρi+1 . By
denoting V0 = supx V (x), this inequality holds if
Proposition 3.2 The homogeneous configurations for which ρi (t) = ρe for all i and t are
linearly stable for the discrete traffic model Eq. 20 if and only if
Proof The perturbations to homogeneous solution are the variables εi (t) = ρi (t) − ρe . The
perturbed system is
εi (t + dt) = ρi (t + dt) − ρe = F (ρi (t), ρi+1 (t), ρi+2 (t), ρi−1 (t)) − ρe
(27)
= α εi (t) + β εi+1 (t) + γ εi+2 (t) + ξ εi−1 (t) + o(LC(εi , εi−1 , εi+1 , εi+2 )),
7
∂F
with α = ∂ρ i
, β = ∂ρ∂Fi+1
, γ = ∂ρ∂F
i+2
and ξ = ∂ρ∂F i−1
at (ρe , ρe , ρe , ρe ). General conditions for the
global stability of the discrete schemes can be obtained for a system of N cells with periodic
boundary conditions. The linear perturbed system is ~ε (t+ dt) = M ~ε (t), with ~ε = T (ε1 , . . . , εN )
and M a sparse matrix with (ξ, α, β, γ) on the diagonal. If M = P DP −1 with D a diagonal
matrix, then ~ε (t) = P Dt/ dt P −1 ~ε (0) → ~0 if all the coefficients of D are less than one excepted
one equal to 1. M is circulant therefore the eigenvectors of M are z(ι0 , ι1 , . . . , ιm−1 ) with
ι = exp i 2πl
N and z ∈ Z, and the eigenvalues are λl = α + βιl + γι2l + ξι−1 l . The system is
linearly stable if |λl | < 1 for all l = 1, . . . , N − 1. This is
Lemma 3.3 The homogeneous configurations are linearly stable for the Godunov-Euler scheme
Eqs. (20-22) if
2τ < T ` dx ρ2e , (29)
and if dt is sufficiently small.
Proof Mixed with the scheme for the density Eq. (20), Godunov/Euler scheme Eq. (22) is
dt τ ρi − ρi−1 ρi+1 − ρi
F1 (ρi , ρi+1 , ρi+2 , ρi−1 ) = ρi + `(ρi+1 − ρi ) + − , (30)
T dx T dx ρ2i−1 ρ2i
1
and for all dt ≥ 0 if τ ≥ Moreover 1 − α > 0 if τ < 12 T ` dx ρ2e while β is positive only
2
2 T ` dx ρe .
2
if τ < T ` dx ρe and the sign of ξ is the one of −τ .
The stability conditions are distinguished according to the sign of τ .
• If τ < 0 and Eq. (31) holds, then f (x) = α(1 − α)x + 2βξx2 is strictly convex and is
maximal on [−1, 1] for x = −1 or x = 1. Therefore the model is stable if f (−1) < f (1);
this is simply −α(1 − α) < α(1 − α) that is always true since α > 0 if (31) holds and
1 − α > 0 on τ < 0. Therefore the system is stable for all τ < 0.
• Several cases have to be distinguished for τ > 0. We assume in the following that Eq. (31)
holds.
– For 0 < τ < 21 T ` dx ρ2e , we have α, 1 − α, β > 0, ξ < 0 and f (x) = α(1 − α)x + 2βξx2
is strictly concave and maximal for x0 = − α(1−α)4βξ > 0. The model is stable if x0 > 1,
this is
T dx 2τ
dt < − . (32)
` (`ρe )2
This condition is more restrictive than Eq. (31).
8
– For 21 T ` dx ρ2e < τ < T ` dx ρ2e , we have α, β > 0, 1 − α, ξ < 0 and f (−1) > f (1)
therefore the model is unstable. More precisely, f maximal for x0 < −1, i.e. the
unstable solution have shortest wavelength if
−2
1 2 2τ
dt < τ − T ` dx ρe . (33)
2 T dx ρe
This condition is also more restrictive than Eq. (31).
– For τ > T ` dx ρ2e , we have α > 0, 1 − α, β, ξ < 0 and the system is unstable for all dt
with shortest wavelength since f (·) strictly convex and f (−1) > f (1).
The stability conditions for the Godunov/Euler splitting scheme Eq. (22) are summarised in
Fig. 3. The same conditions as the continuous macroscopic model are obtained for dx → 0 and
dt → 0 such that dt / dx → 0.
1 2
0 2 T `dxρe T `dxρ2e τ
Figure 3: Summary of the stability conditions for the Godunov/Euler splitting scheme Eq. (22).
Lemma 3.4 The homogeneous configurations are linearly stable for the Godunov-Euler schemes
Eqs. (20-23) and Eqs. (20-24) if
2|τ | < T dx ρe , (34)
and if dt is sufficiently small.
Proof The Godunov numerical schemes Eq. (23) and Eq. (24) are respectively
dt ` τ ρi+1 − ρi+2 ρi − ρi+1
F2 (ρi , ρi+1 , ρi+2 , ρi−1 ) = ρi + ρi+1 − ρi + − , (35)
T dx T dx ρi ρi−1
and
dt ` ρi+1 ρi
F3 (ρi , ρi+1 , ρi+2 , ρi−1 ) = ρi + − . (36)
T dx 1 − T τdx ρ 1 − ρ 1 1 − T
τ
dx ρ
1
− 1
ρ
i+2 i+2 i+1 i
9
• If τ < 0 and Eq. (37) holds, f (x) = β(1 − β)x + 2αγx2 is strictly convex is maximal on
[−1, 1] for x = −1 or x = 1. Therefore the model is stable if f (−1) < f (1); this is
1 T dx
τ > − T dx ρe and dt < . (38)
2 ` + T 2`τ
dx ρe
The condition for dt is weaker than (37) since τ is negative. If τ ≤ − 12 T dx ρe then the
system is unstable at the shortest wave-length frequency. A sufficiently condition for that
the finite system produces the shortest frequency is simply N ≥ 2. Note that no condition
holds on dt if τ ≤ −T dx ρe .
• If τ > 0 and Eq. (37) holds then f (x) = β(1 − β)x + 2αγx2 is concave and is maximum
at arg supx f (x) = x0 = − β(1−β) 2 2 2 2
4αγ > 0. We know that λ0 = α + β + γ − 2αβ + f (1) = 1
(case l = 0). Therefore the model is stable if x0 > 1; this is
1 T dx 2τ
τ < T dx ρe and dt < − . (39)
2 ` `ρe
The condition for dt is stronger than Eq. (37). If τ ≥ 12 T dx ρe then the system is unstable
at the frequency cos−1 (x0 ) that is reachable in the finite system if N > 2π/ cos−1 (x0 ). We
have x0 → 1/2 + T dx ρe /(4τ ) as dt → 0, going from 1 to 1/2 according to τ (long wave).
The stability conditions for the Godunov/Godunov and Godunov schemes Eq. (23) and Eq. (24)
are summarised in Fig. (4). The same conditions as the microscopic model are obtained at the
limit dt → 0 for dx = 1/ρe , i.e. a space step equal to the mean spacing.
e
Shortest Shortest Wavelength
wavelength wavelength from N/2 to N/6
−T dxρe − 12 T dxρe 0 1
2 T dxρe τ
Figure 4: Summary of the stability conditions for the Godunov/Godunov and Godunov schemes
Eq. (23) and Eq. (24). Note that we have the additional condition τ < dx /V0 , with V0 =
supx V (x), for the simple Godunov scheme Eq. (24).
10
positive and bounded if the optimal velocity functions are so defined. Moreover, the density
remains bounded in [1, ρM ], with ρM = 1/`, if
It is easy to check that such property holds only if τ ≤ 0 for the Godunov/Euler Eq. (30), while
it holds for τ ≥ − dx ρe /W 0 with the Godunov/Godunov scheme Eq. (35), and for τ < dx /V0
with the simple Godunov scheme Eq. (36). As the microscopic model, Eq. (35) and Eq. (36) are
able to describe macroscopically unstable homogeneous solutions with large waves by ensuring
that speed and density remain positive and bounded. The relation between instability and self-
sustained traffic waves (or jamiton) are notably described in [17, 34, 35, 41] with microscopic and
macroscopic second order models. In the next section, we analyse by simulation the unstable
solutions we get with the first order models for different initial conditions.
4 Simulation results
In this section numerical simulations of the microscopic model Eq. (5) and of the simple Godunov
scheme Eq. (36) macroscopic model are compared. The car-following model Eq. (5) is simulated
using an explicit Euler scheme. A ring (periodic boundaries) with a length 101 and 50 vehicles
is considered. The optimal speed functions are W (∆) = max{0, min{2, ∆ − 1}} and V (ρ) =
W (1/ρ) corresponding to a triangular fundamental diagram, while the reaction time is τ = 1.
The values of the parameters are set to obtain unstable homogeneous solutions. The time step
is dt = 0.01. The space step for the Godunov scheme is the mean spacing dx = 101/50 = 2.02 in
order to match the stability conditions of both microscopic and macroscopic model (see Eq. (6)
and Fig. 4) and to hold the CFL conditions (see Eq. (25) and Fig. 4). Three experiments are
carried out with different initial conditions. In the first one, the initial configuration is a jam.
The initial condition is random in the second experiment while it is a perturbed homogeneous
configuration in the last one.
4.1 Trajectories
In Figs. 5, 6 and 7, the trajectories of the microscopic model and the time series for the density
by cell for the discrete macroscopic model (gray levels) are plot for respectively the jam, random
and perturbed initial conditions. The jam stationary propagates within the first experiment in
Fig. 5. Both microscopic and macroscopic models rigorously describe the same dynamics. The
dynamics obtained does not perfectly coincide for the random and perturbed initial conditions
(see Figs. 6 and 5). Yet most of the dynamics seems to be well recaptured and notably the
self-sustained emergence of traffic stop-and-go waves. Note that the waves propagate backward
with the speed −`/T that is close to the value empirically observed (see [32]).
11
Jam initial configuration
100
80
60
Time
40
20
0
0 20 40 60 80 100
Space
Figure 5: The trajectories of the microscopic model (cyan curves) and the time series for the
density by cell for the discrete macroscopic model (gray levels) for jam initial conditions.
0 20 40 60 80 100
Space
Figure 6: The trajectories of the microscopic model (cyan curves) and the time series for the
density by cell for the discrete macroscopic model (gray levels) for random initial conditions.
The sequences obtained for the perturbed initial conditions (see Fig. 7) are presented in Fig. 8.
The performances are instantaneous ones in the sense that they correspond to instantaneous
12
Perturbed initial configuration
100
80
60
Time
40
20
0
0 20 40 60 80 100
Space
Figure 7: The trajectories of the microscopic model (cyan curves) and the time series for the
density by cell for the discrete macroscopic model (gray levels) for perturbed initial conditions.
measurements for a vehicle (microscopic model) and a cell (macroscopic) in the system. The
variability in such diagram is larger than the one of the aggregated fundamental diagram plotted
in Fig. 1 where the performances were averaged over time intervals.
Both microscopic and macroscopic systems converge to limit-cycles with self-sustained stop-
and-go waves resulting in hysteresis curves in the microscopic fundamental diagram. Such
phenomenon generate scattering of the fundamental diagram for which some bounds can be
calculated [47, 11, 19, 12, 41, 16]. The bounds V + and V − for the fundamental diagrams can
here intuitively been determined from the microscopic model. The upper bound V + corresponds
to the sequence of a vehicle moving at maximal speed V0 behind a stopped vehicle:
+ ρ
V (ρ) = Ṽ (ρ, 1/`) = V . (42)
1 + τ ρV (ρ)
Due to the reaction time, the distance tends to be smaller and the fundamental diagram is
‘over-estimated’. Oppositely, the lower bound V − corresponds to the sequence of a stopped
vehicle following a predecessor moving at the maximal speed V0 :
− ρ
V (ρ) = Ṽ (ρi , 0) = V . (43)
1 − τ ρ(V0 − V (ρ))
Here the reaction time induces a delay in the acceleration and an under-estimation of the
fundamental diagram.
As in [41, 16], the bounds Eqs. (42) and (43) obtained with the macroscopic model are compared
to real instantaneous pedestrians and road traffic data in Figs. 9 and 10. The pedestrians
data comes from a laboratory experiment with participants in a ring geometry [18]. Several
experiments have been carried out with different density levels. The road traffic data are real
measurement of trajectories on an American highway [33]. The speed, density and the flow
13
Microscopic model Macroscopic model
2.0
2.0
V (·)
Bounds
1.5
1.5
Speed
1.0
1.0
500
0.5
0.5
400
0.0
0.0
300
0.4 0.6 0.8 1.0 0.4 0.6 0.8 1.0
200
0.0 0.2 0.4 0.6 0.8 1.0
100
Flow
0
Time
Density Density
Figure 8: Sequence of speed and flow / density relation for the perturbed initial conditions (see
Fig. 7). Left, for one vehicle (microscopic model) and right, for one cell (macroscopic model).
are measured as previously (i.e. the density is the inverse of the spacing while the flow is the
product of the
density by the speed). A triangular fundamental diagram with 3 parameters
V (ρ) = min V0 , T1 (1/ρ − 1) is used again. The parameters are the ones of an estimation by
least squares for the pedestrians V0 = 0.9 m/s, ` = 0.3 m and T = τ = 1 s, see Fig. 9, while
V0 = 15 m/s, ` = 5 m and T = τ = 2 s for the vehicles, see Fig. 10. The bounds present a
reasonable agreement with the data, even if no clustering of measurements are observed around
them.
5 Conclusion
Starting from a speed following model, we derive a first order convection-diffusion continuum
traffic flow model that we discretised using Godunov and Euler schemes. Simulation results
shown that discrete macroscopic models can recapture the dynamics of the microscopic model,
if specific values for the space discretization are chosen. More precisely, the linear stability
conditions of the homogeneous solutions for the macroscopic models match the ones of the
microscopic model for specific values of the space discretization and sufficiently small time
steps.
For unstable conditions, i.e. for large reaction times, the dynamics obtained describe self-
sustained stop-and-go waves, with hysteresis cycles and a large scattering of the fundamental
flow/density diagram. Such characteristics are observed in real data [44, 24, 23, 9, 48] as well
as for second order models [47, 11, 19, 6, 17, 41, 12, 16]. Here it is achieved with first order
14
1.0
1.0
Speed (m/s)
Flow (1/s)
0.6
0.5
0.2
0.0
-0.2
0 1 2 3 4 0 1 2 3 4
Figure 9: Instantaneous speed/density and flow/density measurements for real pedestrian flows
[18] and the bounds Eqs. (42) and (43) for V0 = 0.9 m/s, ` = 0.3 m and T = τ = 1 s.
15
0.8
Speed (m/s)
Flow (1/s)
10
0.4
5
0.0
0
0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20
Figure 10: Individual speed/density and flow/density measurements for real road traffic flows
[33] and the bounds Eqs. (42) and (43) for V0 = 15 m/s, ` = 5 m and T = τ = 2 s.
models ensuring by construction that the models are physical and ‘collision-free’ (i.e. bounded
and positive speed as well as density). Further investigations are necessary to understand the
impact of the shape of the optimal velocity function on the characteristics of the waves.
The macroscopic model corresponding to the follow-the-leader model is a first order elliptic
convection-diffusion equation, for which the convection part is calibrated by the optimal velocity
function (i.e. the fundamental diagram), while the diffusion is proportional to the reaction time
parameter. More precisely the diffusion is negative in deceleration phases where the density get
higher, and it is positive in acceleration phases where the density decreases. Such mechanism
seems to be responsible for the appearance of oscillations and self-sustained non-linear stop-
and-go waves in the system. This observation remains to be confirmed rigorously, yet it could
give us a way to explain the wave formations.
15
References
[1] C. Appert-Rolland, P. Degond, and S. Motsch. Two-way multi-lane traffic model for pedes-
trians in corridors. Netw. Heterog. Media, 6(3):351–381, 2011. 1
[2] A. Aw, A. Klar, M. Rascle, and T. Materne. Derivation of continuum traffic flow mod-
els from microscopic follow-the-leader models. SIAM Journal on Applied Mathematics,
63(1):259–278, 2002. 4, 7
[3] A. Aw and M. Rascle. Resurrection of ”second order” models of traffic flow. SIAM Journal
on Applied Mathematics, 60(3):916–938, 2000. 3
[6] F. Berthelin, P. Degond, M. Delitala, and M. Rascle. A model for the formation and
evolution of traffic jams. Archive for Rational Mechanics and Analysis, 187(2):185–220,
2008. 3, 14
[7] Sebastien Blandin, Dan Work, Paola Goatin, Benedetto Piccoli, and Alexandre Bayen. A
general phase transition model for vehicular traffic. SIAM journal on Applied Mathematics,
71(1):107–127, 2011. 3
[8] R. Bürger and K. H. Karlsen. On a diffusively corrected kinematic-wave traffic flow model
with changing road surface conditions. Mathematical Models and Methods in Applied Sci-
ences, 13(12):1767–1799, 2003. 6
[11] R.M. Colombo. Hyperbolic phase transitions in traffic flow. SIAM Journal on Applied
Mathematics, 63(2):708–721, 2003. 3, 13, 14
[12] R.M. Colombo, F. Marcellini, and M. Rascle. A 2-phase traffic model based on a speed
bound. SIAM Journal on Applied Mathematics, 70(7/8):2652–2666, 2010. 3, 13, 14
[13] C. F. Daganzo. Requiem for second-order fluid approximations of traffic flow. Transport.
Res. B: Meth., 29(4):277–286, 1995. 2, 7
[14] L. C. Davis. Modifications of the optimal velocity traffic model to include delay due to
driver reaction time. Physica A, 319(0):557–567, 2003. 2
[15] L.C. Edie. Discussion of traffic stream measurements and definitions. In J. Almond,
editor, Proc. of the 2nd International Symposium on Transportation and Traffic Theory,
pages 139–154, 1963. 11
[16] S. Fan, M. Herty, and B. Seibold. Comparative model accuracy of a data-fitted generalized
Aw-Rascle-Zhang model. Netw. Heterog. Media, 9(2):239–268, 2014. 3, 13, 14
16
[17] M.R. Flynn, A.R. Kasimov, J.-C. Nave, R.R. Rosales, and B. Seibold. Self-sustained
nonlinear waves in traffic flow. Phys. Rev. E, 79(5):056113, 2009. 3, 11, 14
[19] P. Goatin. The Aw–Rascle vehicular traffic flow model with phase transitions. Mathematical
and Computer Modelling, 44(34):287–303, 2006. 3, 13, 14
[21] J.M. Greenberg. Extensions and amplifications of a traffic model of Aw and Rascle. SIAM
Journal on Applied Mathematics, 62(3):729–745, 2002. 3
[24] B. S. Kerner and H. Rehborn. Experimental properties of phase transitions in traffic flow.
Phys. Rev. Lett., 79:4030–4033, 1997. 2, 14
[25] J.-P. Lebacque. Les modèles macroscopiques du trafic. Annales des Ponts, 67:24–45, 1993.
1
[26] J.-P. Lebacque. The Godunov scheme and what it means for first order traffic flow models.
In J.-B. Lesort, editor, Proc. of the 13th International Symposium on Transportation and
Traffic Theory, pages 647–677, 1996. 7
[27] J.-P. Lebacque, S. Mammar, and H. H. Salem. Generic second order traffic flow modelling.
In Proceedings of the 17th Symposium on Transportation and Traffic Theory, pages 755–
776. Elsevier, Oxford, 2007. 3
[28] Jean-Patrick Lebacque, H Haj-Salem, and Salim Mammar. Second order traffic flow mod-
eling: supply-demand analysis of the inhomogeneous Riemann problem and of boundary
conditions. In Proceedings of the 10th Euro Working Group on Transportation (EWGT),
2005. 3
[29] M. H. Lighthill and G. B. Whitham. On kinematic waves II: a theory of traffic flow on
long, crowded roads. In Proceedings of the Royal Society of London series A, volume 229,
pages 317–345, 1955. 2
[30] P. Nelson. Synchronized traffic flow from a modified Lighthill–Whitman model. Phys. Rev.
E, 61:R6052–R6055, 2000. 6
[31] G. F. Newell. Nonlinear effects in the dynamics of car-following. Op. Res., 9(2):209–229,
1961. 2
[32] K. Nishinari, M. Treiber, and D. Helbing. Interpreting the wide scattering of synchronized
traffic data by time gap statistics. Phys. Rev. E, 68:067101, 2003. 11
17
[33] US Department of Transportation. NGSIM: Next Generation Simulation — http://www.
ngsim.fhwa.dot.gov. 13, 15
[34] G. Orosz and G. Stépán. Subcritical Hopf bifurcations in a car-following model with
reaction-time delay. Proc. Roy. Soc. London Ser. A, 462(2073):2643–2670, 2006. 2, 11
[35] G. Orosz, R. E. Wilson, and G. Stépán. Traffic jams: dynamics and control. Proc. Roy.
Soc. London Ser. A, 368(1928):4455–4479, 2010. 2, 11
[36] G. Orosz, R. E. Wilson, R. Szalai, and G. Stépán. Exciting traffic jams: Nonlinear phe-
nomena behind traffic jam formation on highways. Phys. Rev. E, 80(4):046205, 2009. 2
[37] H. J. Payne. Models of freeway traffic and control. In Mathematical Models of Public
Systems, Simulation Council Proceedings Series, volume 1, pages 51–61, 1971. 2
[40] A. Schadschneider and A. Seyfried. Empirical results for pedestrian dynamics and their
implications for modeling. Netw. Heterog. Media, 6(3):545–560, 2011. 1
[41] B. Seibold, M.R. Flynn, A.R. Kasimov, and R.R. Rosales. Constructing set-valued fun-
damental diagrams from jamiton solutions in second order traffic models. Netw. Heterog.
Media, 8(3):745–772, 2013. 3, 11, 13, 14
[42] A. Tordeux and A. Seyfried. Collision-free nonuniform dynamics within continuous optimal
velocity models. Phys. Rev. E, 90:042812, 2014. 3
[43] M. Treiber and A. Kesting. Traffic Flow Dynamics. Springer, Berlin, 2013. 1, 2
[44] J. Treiterer and J.A. Myers. The hysteresis phenomenon in traffic flow. In D. J. Buckley, ed-
itor, Transportation and Traffic Theory, Proceedings of the Sixth International Symposium,
pages 13–38, 1974. 2, 14
[45] J.H. Wardrop. Some theoretical aspects of road traffic research. Proceedings of the Insti-
tution of Civil Engineers, 1(3):325–362, 1952. 11
[46] G.B. Whitham. Linear and Nonlinear Waves. Pure and Applied Mathematics: A Wiley
Series of Texts, Monographs and Tracts. Wiley, 2011. 2
[47] H.M. Zhang. A non-equilibrium traffic model devoid of gas-like behavior. Transport. Res.
B: Meth., 36(3):275–290, 2002. 3, 13, 14
18
Contents
1 Introduction 1
4 Simulation results 11
4.1 Trajectories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
4.2 Fundamental diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
5 Conclusion 14
References 16
19