Second order traffic flow models on networks

Second Order Traffic Flow Models on Networks
Guillaume Costeseque
in collaboration with J-P. Lebacque (IFSTTAR)
Inria Sophia-Antipolis Méditerranée
LJAD seminar, UNS
January 12, 2017
G. Costeseque (Inria) GSOM on networks Nice, Jan. 12 2017 1 / 58

Motivation
Traﬃc ﬂows on a network
[Caltrans, Oct. 7, 2015]

Motivation
Traﬃc ﬂows on a network
[Caltrans, Oct. 7, 2015]
Road network ≡ graph made of edges and vertices

Motivation
Breakthrough in traffic monitoring
Traffic monitoring
“old”: loop detectors at fixed locations (Eulerian)
“new”: GPS devices moving within the traffic (Lagrangian)
Data assimilation of Floating Car Data
[Mobile Millenium, 2008]

Motivation
Outline
1 Introduction to traﬃc
2 Variational principle applied to GSOM models
3 GSOM models on a junction

Introduction to traﬃc
Outline

Introduction to traﬃc Macroscopic models
Convention for vehicle labeling
N
x
t
Flow

Three representations of traﬃc ﬂow
Moskowitz’ surface
Flow
x
t
N
x
See also [Makigami et al, 1971], [Laval and Leclercq, 2013]

Notations: macroscopic
N(t, x) vehicle label at (t, x)
the ﬂow Q(t, x) = lim
∆t→0
N(t + ∆t, x) − N(t, x)
∆t
= ∂tN(t, x)
x
N(x, t ± ∆t)
the density ρ(t, x) = lim
∆x→0
N(t, x) − N(t, x + ∆x)
∆x
= −∂x N(t, x)
x
∆x
N(x ± ∆x, t)
the stream speed (mean spatial speed) V (t, x).

Macroscopic models
Hydrodynamics analogy
Two main categories: first and second order models
Two common equations:



∂tρ(t, x) + ∂x Q(t, x) = 0 conservation equation
Q(t, x) = ρ(t, x)V (t, x) definition of flow speed
(1)
x x + ∆x
ρ(x, t)∆x
Q(x, t)∆t Q(x + ∆x, t)∆t

Introduction to traﬃc Focus on LWR model
First order: the LWR model
LWR model [Lighthill and Whitham, 1955], [Richards, 1956] [6, 7]
Scalar one dimensional conservation law
∂tρ(t, x) + ∂x F (ρ(t, x)) = 0 (2)
with
F : ρ(t, x) → F (ρ(t, x)) := Q(t, x)

Overview: conservation laws (CL) / Hamilton-Jacobi (HJ)
Eulerian Lagrangian
t − x t − n
CL
Variable Density ρ Spacing r
Equation ∂tρ + ∂x F(ρ) = 0 ∂tr + ∂nV (r) = 0
HJ
Variable Label N Position X
N(t, x) =
+∞
x
ρ(t, ξ)dξ X(t, n) =
+∞
n
r(t, η)dη
Equation ∂tN + H (∂x N) = 0 ∂tX + V (∂nX) = 0
Hamiltonian H(p) = −F(−p) V(p) = −V (−p)

Fundamental diagram (FD)
Flow-density fundamental diagram F
Empirical function with
ρmax the maximal or jam density,
ρc the critical density
Flux is increasing for ρ ≤ ρc: free-ﬂow phase
Flux is decreasing for ρ ≥ ρc: congestion phase
ρmax
Density, ρ
ρmax
Density, ρ
0
Flow, F
0
Flow, F
0 ρmax
Flow, F
Density, ρ
[Garavello and Piccoli, 2006]

Introduction to traﬃc Second order models
Motivation for higher order models
Experimental evidences
fundamental diagram: multi-valued in congested case
[S. Fan, U. Illinois], NGSIM dataset

Motivation for higher order models
Experimental evidences
fundamental diagram: multi-valued in congested case
phenomena not accounted for: bounded acceleration, capacity drop...
Need for models able to integrate measurements of diﬀerent traﬃc
quantities (acceleration, fuel consumption, noise)

GSOM family [Lebacque, Mammar, Haj-Salem 2007] [5]
Generic Second Order Models (GSOM) family



∂tρ + ∂x (ρv) = 0 Conservation of vehicles,
∂t(ρI) + ∂x (ρvI) = ρϕ(I) Dynamics of the driver attribute I,
v = I(ρ, I) Speed-density fundamental diagram,
(3)
Speciﬁc driver attribute I
the driver aggressiveness,
the driver origin/destination or path,
the vehicle class,
...
Flow-density fundamental diagram
F : (ρ, I) → ρI(ρ, I).




∂tI + v∂x I = ϕ(I) Dynamics of the driver attribute I,
(3)
the vehicle class,
...
F : (ρ, I) → ρI(ρ, I).




∂tI + v∂x I = 0 Dynamics of the driver attribute I,
(3)
the vehicle class,
...
F : (ρ, I) → ρI(ρ, I).

(continued)
Kinematic waves or 1-waves:
similar to the seminal LWR model
density variations at speed ν = ∂ρI(ρ, I)
driver attribute I is continuous
Contact discontinuities or 2-waves:
variations of driver attribute I at speed ν = I(ρ, I)
the ﬂow speed v is constant.

Examples of GSOM models
• LWR model = GSOM model with no speciﬁc driver attribute
• LWR model with bounded acceleration = GSOM model with I := v
• ARZ model = GSOM with I := v + p(ρ)



∂tρ + ∂x (ρv) = 0,
∂t(ρw) + ∂x (ρvw) = 0,
w = v + p(ρ)
• Generalized ARZ model [Fan, Herty, Seibold]
• Multi-commodity models (multi-class, multi-lanes) of [Jin and
Zhang], [Bagnerini and Rascle] or [Herty, Kirchner, Moutari and
Rascle], [Klar, Greenberg and Rascle]
• Colombo 1-phase model
• Stochastic GSOM model [Khoshyaran and Lebacque]

Variational principle applied to GSOM models
Outline

Variational principle applied to GSOM models LWR model
LWR in Eulerian (t, x)
Cumulative vehicles count (CVC) or Moskowitz surface N(t, x)
Q = ∂tN and ρ = −∂x N
If density ρ satisfies the scalar (LWR) conservation law
∂tρ + ∂x F(ρ) = 0
Then N satisfies the first order Hamilton-Jacobi equation
∂tN − F(−∂x N) = 0 (4)

Legendre-Fenchel transform with F concave (relative capacity)
M(q) = sup
ρ
[F(ρ) − ρq]
M(q)
u
w
Density ρ
q
q
Flow F
w u
q
Transform M
−wρmax

(continued)
Lax-Hopf formula (representation formula) [Daganzo, 2006]
N(T, xT ) = min
u(.),(t0,x0)
T
t0
M(u(τ))dτ + N(t0, x0),
˙X = u
u ∈ U
X(t0) = x0, X(T) = xT
(t0, x0) ∈ J
(5) Time
Space
J
(T, xT )˙X(τ)
(t0, x0)
Viability theory [Claudel and Bayen, 2010]

(Historical note)
Dynamic programming [Daganzo, 2006] for triangular FD
(u and w free and congested speeds)
Flow, F
w
u
0 ρmax
Density, ρ
u
x
w
t
Time
Space
(t, x)
Minimum principle [Newell, 1993]
N(t, x) = min N t −
x − xu
u
, xu ,
N t −
x − xw
w
, xw + ρmax (xw − x) ,
(6)

LWR in Lagrangian (n, t)
Consider X(t, n) the location of vehicle n at time t ≥ 0
v = ∂tX and r = −∂nX
If the spacing r := 1/ρ satisfies the LWR model (Lagrangian coord.)
∂tr + ∂nV(r) = 0
Then X satisfies the first order Hamilton-Jacobi equation
∂tX − V(−∂nX) = 0. (7)

LWR in Lagrangian (n, t)
(continued)
Dynamic programming for triangular FD
1/ρcrit
Speed, V
u
−wρmax
Spacing, r
1/ρmax
−wρmax
n
t
(t, n)
Time
Label
Minimum principle ⇒ car following model [Newell, 2002]
X(t, n) = min X(t0, n) + u(t − t0),
X(t0, n + wρmax (t − t0)) + w(t − t0) .
(8)

Variational principle applied to GSOM models GSOM family
GSOM in Lagrangian (n, t)
From [Lebacque and Khoshyaran, 2013], GSOM in Lagrangian



∂tr + ∂N v = 0 Conservation of vehicles,
∂tI = 0 Dynamics of I,
v = W(N, r, t) := V(r, I(N, t)) Fundamental diagram.
(9)
Position X(N, t) :=
t
−∞
v(N, τ)dτ satisﬁes the HJ equation
∂tX − W(N, −∂N X, t) = 0, (10)
And I(N, t) solves the ODE
∂tI(N, t) = 0,
I(N, 0) = i0(N), for any N.

(continued)
Legendre-Fenchel transform of W according to r
M(N, c, t) = sup
r∈R
{W(N, r, t) − cr}
M(N, p, t)
pq
W(N, q, t)
W(N, r, t)
q r
p
p
u
c
Transform M

(continued)
Lax-Hopf formula
X(NT , T) = min
u(.),(N0,t0)
T
t0
M(N, u, t)dt + c(N0, t0),
˙N = u
u ∈ U
N(t0) = N0, N(T) = NT
(N0, t0) ∈ K
(11)

(continued)
Optimal trajectories = characteristics
˙N = ∂r W(N, r, t),
˙r = −∂N W(N, r, t),
(12)
System of ODEs to solve
Diﬃculty: not straight lines in the general case

Variational principle applied to GSOM models Methodology
General ideas
First key element: Lax-Hopf formula
Computations only for the characteristics
X(NT , T) = min
(N0,r0,t0)
T
t0
M(N, ∂r W(N, r, t), t)dt + c(N0, r0, t0),
˙N(t) = ∂r W(N, r, t)
˙r(t) = −∂N W(N, r, t)
N(t0) = N0, r(t0) = r0, N(T) = NT
(N0, r0, t0) ∈ K
(13)
K := Dom(c) is the set of initial/boundary values

General ideas
(continued)
Second key element: inf-morphism prop. [Aubin et al, 2011]
Consider a union of sets (initial and boundary conditions)
K =
l
Kl ,
then the global minimum is
X(NT , T) = min
l
Xl (NT , T), (14)
with Xl partial solution to sub-problem Kl .

IBVP
Consider piecewise affine initial and boundary conditions:
• initial condition at time t = t0 = initial position of vehicles ξ(·, t0)
• “upstream” boundary condition = trajectory ξ(N0, ·) of the first
vehicle,
• and internal boundary conditions given for instance by cumulative
vehicle counts at fixed location X = x0.

Variational principle applied to GSOM models Numerical example
Fundamental Diagram and Driver Attribute
0 20 40 60 80 100 120 140 160 180 200
0
500
1000
1500
2000
2500
3000
3500
Density ρ (veh/km)
FlowF(veh/h)
Fundamental diagram F(ρ,I)
0 5 10 15 20 25 30 35 40 45 50
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
Label N
Initial conditions I(N,t
0
)
DriverattributeI
0

Initial and Boundaries Conditions
0 5 10 15 20 25 30 35 40 45 50
5
10
15
20
25
30
35
40
45
50
Label N
Initial conditions r(N,t
0
)
Spacingr0
(m)
0 5 10 15 20 25 30 35 40 45 50
−1400
−1200
−1000
−800
−600
−400
−200
0
Label N
PositionX(m)
Initial positions X(N,t
0
)
0 20 40 60 80 100 120
12
14
16
18
20
22
24
26
28
30
Time t (s)
Spacing r(N
0
,t)
Spacingr
0
(m.s
−1
)
0 20 40 60 80 100 120
0
500
1000
1500
2000
2500
Time t (s)
PositionX0
(m)
Position X(N
0
,t)

Numerical result (Initial cond. + ﬁrst traj.)

Numerical result (Initial cond. + ﬁrst traj.)
0 20 40 60 80 100 120
−1500
−1000
−500
0
500
1000
1500
2000
2500
Location(m)
Time (s)
Vehicles trajectories

Numerical result (Initial cond.+ 3 traj.)

Numerical result (Initial cond. + 3 traj.)
0 20 40 60 80 100 120
−1500
−1000
−500
0
500
1000
1500
2000
2500
Location(m)
Time (s)
Vehicles trajectories

Numerical result (Initial cond. + 3 traj. + Eulerian data)

GSOM models on a junction
Outline

GSOM models on a junction
The whole picture
We need
(i) a link model
(ii) a junction model
(iii) the upstream (resp. downstream) boundary conditions for an
incoming (resp. outgoing) link
(iv) link-node and node-link interfaces

GSOM models on a junction Recalls
GSOM lagrangian
General expressions of GSOM family
In Eulerian,



∂t(ρI) + ∂x (ρvI) = ρϕ(I) Dynamics of the driver attribute I,
v = I(ρ, I) Fundamental diagram,
(15)
Transformed in Lagrangian,



∂T r + ∂nv = 0 Conservation of vehicles,
∂T I = ϕ(I) Dynamics of the driver attribute I,
v = V(r, I) Fundamental diagram.
(16)

GSOM lagrangian
Following classical approach [3, 4] we set
∆t, ∆N time and particle steps;
rt
n := r(t∆t, n∆N), for any t ∈ N and any n ∈ Z
and It
n := I(t∆t, n∆N).
Numerical scheme



rt+1
n := rt
n +
∆t
∆N
V t
n−1 − V t
n ,
V t
n := V (rt
n, It
n ) ,
It+1
n = It
n + ∆tϕ (It
n )
(17)
CFL condition:
∆N
∆t
≥ sup
N,r,t
|∂r V(r, I(t, N))| . (18)

GSOM lagrangian (HJ)
Introduce X(T, N) the position of particle N at time T and satisfying
r = −∂NX and v = ∂T X
such that
∂T X = V (−∂NX, I) ,
∂T I = ϕ(I).
(19)
Numerical scheme for HJ equation



Xt+1
n = Xt
n + ∆t V t
n ,
V t
n := V
Xt
n−1 − Xt
n
∆N
, It
n ,
It+1
n = It
n + ∆t ϕ (It
n)
(20)

Boundary conditions
We have two diﬀerent solutions:
“Classical” supply-demand methodology [3, 2, 1]
but it implies to work with ﬂows;
Using tools developed in [Lebacque, Khoshyaran, (2013)] [4] that
allow to compute directly spacing.

GSOM models on a junction Downstream boundary conditions
Downstream boundary conditions
(Continued)
tn ∆t
X
tn−1
n
∆t
tn−1 ∆t
V(., I)
σt
r∗
r∗
xS
rcrit(I)
rt
n∆N
Xt
n Xt
n−1
r
x
t
xS
x
(n) (n − 1)
(n)
(n − 1)
X
tn−1
n−1
Exit point S located at xS
Boundary data = downstream supply
σt = σ(t∆t).
(n) the last particle located on the link
(or at least a fraction η∆N of it is still
on the link, with 0 ≤ η < 1).

tn ∆t
X
tn−1
n
∆t
tn−1 ∆t
V(., I)
σt
r∗
r∗
xS
rcrit(I)
rt
n∆N
Xt
n Xt
n−1
r
x
t
xS
x
(n) (n − 1)
(n)
(n − 1)
X
tn−1
n−1

(Continued)
Computational steps:
1 Deﬁne the spacing associated to particle (n) as rt
n :=
Xt
n−1 − Xt
n
∆N

(Continued)
n :=
Xt
n−1 − Xt
n
∆N
2 Deﬁne the proportion of (n) already out η :=
xS − Xt
n
rt
n∆N

(Continued)
n :=
Xt
n−1 − Xt
n
∆N
xS − Xt
n
rt
n∆N
3 Distinguish two cases:
• either V(rt
n, It
n ) ≤ σt
rt
n: spacing is conserved.
• or V(rt
n, It
n ) > σt
rt
n: then, we solve V(rt
n, It
n ) = σt
rt
n
and choice of the smallest value rt
n ← r∗ (congested)

(Continued)
n :=
Xt
n−1 − Xt
n
∆N
xS − Xt
n
rt
n∆N
3 Distinguish two cases:
• either V(rt
n, It
n ) ≤ σt
rt
n: spacing is conserved.
• or V(rt
n, It
n ) > σt
rt
n: then, we solve V(rt
n, It
n ) = σt
rt
n
and choice of the smallest value rt
n ← r∗ (congested)
4 Update Xt+1
n (Euler scheme)
• If Xt+1
n > xS , go to next particle n ← n + 1
• Else, update η ← η −
∆t
rt
n∆N
V (rt
n, It
n ).

GSOM models on a junction Upstream boundary conditions
Upstream boundary conditions
(Continued)
tn+1
(t + 1)∆t
(t − 1)∆t
tn
r
δt
xE Xt
n
r∗ r∗
V(., I)
rcrit(I)
rt
n+1∆N
Xt
n+1
(n + 1) (n)
x
x
t
xE
∆t
(n + 1)
(n)
t∆t
Entry point E located at xE
Boundary data = (discrete) upstream
demand δt = δ(t∆t)
n the last vehicle entered in the link
next particle (n + 1) is still part of the
demand and will enter in the link at
time (t + ε)∆t

We don’t know the position of next particle!
εn+1∆t
tn
tn+1
x
xE Xt
n
δt
Xt
n+1
qt
ηrt
n+1∆N
(n + 1) (n)
x
xE
t
(t − 1)∆t
(n)(n + 1)
(t + 1)∆t
t∆t
∆t
σt
loc

(Continued)
1 Instantiation:
We initialize the fraction η
η = qt−1 (t∆t − tn)
∆N
and
rt
n+1 =
Xt
n − xE
η∆N
.
We introduce the local supply
σt
loc = Ξ
1
rt
n+1
, It
n+1, It
n ; xE for any t ∈ N, n ∈ Z,
Let Ft
be the number of particles stored inside the upstream “queue”.

(Continued)
2 Stock model: The evolution of the stock Ft is given by
Ft+1
= Ft
+ (δt
− qt
)∆t, (21)
where δt is the (cumulative) demand and qt is the eﬀective inﬂow.
• if Ft
> 0, then there is a (vertical) queue upstream and
qt
= min σt
loc , Qmax(It
n+1) ,
Ft
∆t
+ δt
,
• if Ft
= 0, then there is no queue and
qt
= min σt
loc , δt
.

(Continued)
3 Update: Particle (n + 1) is generated if and only if
η∆N + qt∆t ≥ ∆N.
if qt
∆t < (1 − η)∆N, then
η ← η +
qt
∆t
(1 − η)∆N
.
if qt
∆t ≥ (1 − η)∆N, then the particle (n + 1) has entered the link at
time tn+1 = (t + εn+1)∆t where
εn+1 =
(1 − η)∆N
qt ∆t
.
The position of particle (n + 1) is updated
Xt+1
n+1 = xE + (1 − εn+1)∆t V rt
n+1, It
n+1 .
Go to next particle n ← n + 1.

(Continued)
4 Final update: We compute the attribute
It+1
n+1 = It
n+1 + ∆t ϕ It
n
and update the time step t ← t + 1.

GSOM models on a junction Junction model
Junction model
Internal state model (acts like a buﬀer)
Qi
δi(i)
(j)
σjRj
Σi(t)
γij
∆j(t)
[Nz(t), Nz,j(t)]

GSOM models on a junction Junction model
Assignment of particles through the junction
3 methods:
The assignment of particles is known: ∃ (γij )i,j that describe the
proportion of particles coming from any road i ∈ I that want to exit
the junction on road j ∈ J
The path through the junction of each particle n ∈ Z is known:
included in the particle attribute I(t, n) and does not evolve in time
[straightforward]
The origin-destination (OD) information for each particle is known
(may depend on time): consider a reactive assignment model that
give us the path followed by particles.

References
Some references I
M. M. Khoshyaran and J.-P. Lebacque, Lagrangian modelling of
intersections for the GSOM generic macroscopic traffic flow model, in Proceedings
of the 10th International Conference on Application of Advanced Technologies in
Transportation (AATT2008), Athens, Greece, 2008.
J. Lebacque, S. Mammar, and H. Haj-Salem, An intersection model based
on the GSOM model, in Proceedings of the 17th World Congress, The
International Federation of Automatic Control, Seoul, Korea, 2008, pp. 7148–7153.
J.-P. Lebacque, H. Haj-Salem, and S. Mammar, Second order traffic flow
modeling: supply-demand analysis of the inhomogeneous Riemann problem and of
boundary conditions, Proceedings of the 10th Euro Working Group on
Transportation (EWGT), 3 (2005).
J.-P. Lebacque and M. M. Khoshyaran, A variational formulation for higher
order macroscopic traffic flow models of the GSOM family, Procedia-Social and
Behavioral Sciences, 80 (2013), pp. 370–394.

References
Some references II
J.-P. Lebacque, S. Mammar, and H. H. Salem, Generic second order traffic
flow modelling, in Transportation and Traffic Theory 2007. Papers Selected for
Presentation at ISTTT17, 2007.
M. J. Lighthill and G. B. Whitham, On kinematic waves II. A theory of
traffic flow on long crowded roads, Proceedings of the Royal Society of London.
Series A. Mathematical and Physical Sciences, 229 (1955), pp. 317–345.
P. I. Richards, Shock waves on the highway, Operations research, 4 (1956),
pp. 42–51.

References
Thanks for your attention
Any question?
guillaume.costeseque@inria.fr

Second order traffic flow models on networks

More Related Content

What's hot (19)

Similar to Second order traffic flow models on networks (20)

More from Guillaume Costeseque (20)

Recently uploaded (20)

Second order traffic flow models on networks