Traffic Flow Dynamics 6.

The Lighthill-Whitham-Richards (LWR) Model

Lecture 06: The Lighthill-Whitham-Richards (LWR) Model

I 6.1 General: Motivation and Equations

I 6.2 Wave Velocity
I 6.3 Shock-Wave Fronts
I 6.4 Triangular Fundamental Diagram
I 6.5 Unique Properties of the Triangular FD
I 6.6 Bottlenecks
I 6.7 Traffic Lights: Examples of Temporary
bottlenecks
I 6.8 Calculated Examples
I 6.9 Numerics
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6..1 General: Motivation and Equations

6.1 General: Motivation and Equations

I We have three main macroscopic quantities: density ρ, flow Q, and local speed V .
I There is always the static hydrodynamic relation between these quantities arising
directly by the definitions of ρ, Q, and V :

Q = ρV

I Furthermore, vehicle conservation implies the dynamic continuity equation, e.g., for
a homogeneous road:
∂ρ ∂Q ∂ρ ∂(ρV )
+ = + =0
∂t ∂x ∂t ∂x
So, two model-independent relations between the three quantities are always there. To
make a macroscopic flow model that can be simulated, we need a third equation.
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6..1 General: Motivation and Equations

LWR Equations

There are two basic possibilities to specify the missing third relation:
I First-order models or Lighthill-Whitham-Richards (LWR) models specify an
additional static traffic stream model/relation between density and flow,
I Second-order models define a second dynamical equation for the speed

Inserting the fundamental diagram

Q(x, t) = Qe (ρ(x, t))
as traffic stream relation into the continuity equation gives (for homogeneous roads) the
class of first-order models known as LWR models:
∂ρ ∂
∂ρ ∂ρ
+ Qe (ρ) = + Q0e (ρ) = 0 LWR Model
∂t ∂x ∂t ∂x
? Show that the LWR model can also be written as
∂ρ
∂ρ
+ Ve + ρVe0 (ρ) = 0.
∂t ∂x
Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6..1 General: Motivation and Equations

The earliest fundamental diagram of Greenshields

 
ρ
Q(ρ) = V0 ρ 1 − ρmax
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.2 LWR Wave Velocity

6.2 LWR Wave Velocity

Wave ansatz for solving the LWR equa-

∂ Qe (ρ)
tion ∂ρ
∂t + ∂x = 0:

ρ(x, t) = ρ0 (x − ct)

∂ρ
= ρ00 (x − ct) (−c)
∂t
∂Qe (ρ)
= Q0e (ρ) ρ00 (x − ct)
∂x

This solves the LWR equation for all x and t iff −c + Q0e (ρ) = 0 or

c = Q0e (ρ) wave speed of small changes

? The wave speed is never larger than the vehicle speed: c = Q0e (ρ) = V + ρVe0 (ρ). Why? base your
answer on plausibility criteria
! Since there are only interactions front-back but not vice versa
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.2 LWR Wave Velocity

General dynamics of the LWR model

The density dependent wave speed c = Q0e (ρ) means that the density can be imagined as
layers (as in a 3d printer) independently gliding over each other until a shock is formed
where the solution breaks down.
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.3. Formation of Shock Waves

6.3. Formation of Shock Waves

Both the wave and

the wave equation
break down!
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.3. Formation of Shock Waves

Derivation of the shock-wave propagation velocity

I Total vehicle number: n = ρ1 x12 + ρ2 (L − x12 )

I Rate of change as a function of the in- and outflows: dn
dt = Q1 − Q2
I Rate of change as partial time derivative (watch out for the moving boundary with
dc12
dt = c12 ):
dn d

= ρ1 x12 + ρ2 (L − x12 )
dt dt
dx12
= (ρ1 − ρ2 ) = (ρ1 − ρ2 )c12 ⇒
dt
Q2 − Q1 Qe (ρ2 ) − Qe (ρ1 )
c12 = = Shock-wave equation
ρ2 − ρ1 ρ2 − ρ1
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.3. Formation of Shock Waves

6.3 Problems

? Show that, in the case of a triangular fundamental diagram, the wave velocity is
either equal to the vehicle speed or a constant negative value while the shock-wave
propagation velocity can also takes on any value in between.
! Triangular FD: Q(ρ) = min(Qf (ρ), Qc (ρ));
Free traffic: Qf (ρ) = V0 ρ, cf = Q0f (ρ) = V0 = const. (left slope);
Congested traffic: Qc (ρ) = 1/T (1 − ρ/ρmax ), w = Q0c (ρ) = −1/(T ρmax ) = const. (right slope);
Shock velocity c12 : Slope of any line connecting the free with the congested side of the triangle, so
c ≤ c12 ≤ cf

? What is the range of shock propagation velocities in the parabolic fundamental

diagram of Greenshields?
! Greenshields FD: Q(ρ) = V0 ρ(1 − ρ/ρmax ), Q0 (ρ) = V0 (1 − 2ρ/ρmax )
⇒ both the wave and the shock velocities can take on values between Q0 (ρmax ) = −V0 and Q0 (0) = +V0
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.4 Triangular FD

6.4 Triangular FD: Definition and Basic Properties


V0 ρ if ρ ≤ ρc
Qe (ρ) = 1
T (1 − ρleff ) if ρc < ρ ≤ ρmax

I Critical density: ρc = 1
V0 T +leff
I Maximum flow: Qmax = V0 TV+l0
eff

I Maximum 1
density: ρmax = leff

Model parameters:
I Desired speed V0
I Effective vehicle length leff or maximum density ρmax = 1
leff
I Effective time gap T or wave speed w = − lTeff
Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.4 Triangular FD

Typical parameters in different situations

Pedestrian
Parameter Highway City Traffic
Single File
Desired speed V0 120 km/h 50 km/h 1.2 m/s
Time gap T 1.4 s 1.2 s 1.0 s
Max. density ρmax 120 veh/km 120 veh/km 1.5 peds/m
Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.4 Triangular FD

Capacity as a function of the model parameters

 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.5 Properties of the Triangular FD

6.5 Properties of the Triangular


FD
1 − ρleff
Q(ρ) = min V0 ρ,
T

I Analytic expression for the density at capacity: ρc = 1

V0 T +leff

I Analytic expression for the capacity: Qmax = V0 ρc = V TV+l

0
0 eff

V0 free
I Fixed wave propagation velocities: c(Q) =
w = − lTeff congested
I Analytic inverse relations:
Q 1 − QT
ρfree (Q) = , ρcong (Q) = = ρmax (1 − QT )
V0 leff
I By means of the relations Qmax = V0 /(V0 T + leff ) and w = −leff /T , the unobservable
quantities leff and T can be eliminated and the FD reformulated in terms of the observable
parameters V0 , Qmax and w:

if ρ ≤ ρc = QVmax
(
V0 ρ h 0
Qe (ρ) = i
Qmax 1 − Vw0 + wρ if ρ > ρc
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.5 Properties of the Triangular FD

Unique Properties of the Triangular FD (2)

In the triangular FD, waves in one regime (free or congested) remain unchanged
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.5 Properties of the Triangular FD

Unique Properties of the Triangular FD (3)

The upstream jam front free → congested can be calculated by a time delayed ODE
without solving the whole PDE using boundary conditions (e.g., from a detector) at both
ends:
dx12 Q1 (t − τf ) − Q2 (t − τc )
c12 = =
dt ρ1 (t − τf ) − ρ2 (t − τc )

I Q1 (t): free traffic inflow from an upstream stationary detector

I Q2 (t): congested outflow from a downstream stationary detector
I τf = (x12 − x1 )/V0 > 0:
signal travel time from the upstream boundary to the front
I τc = (x12 − xdown )/w > 0:
signal travel time from the downstream boundary to the front

The downstream jam front is either fixed at a bottleneck or moves upstream at velocity
w
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.5 Properties of the Triangular FD

Application: State Detection and Short-Term Forecast: Upstream Jam Front

Both times τf and τc are positive ⇒ real prediction based on vehicle conservation!
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.5 Properties of the Triangular FD

Application to three real situations (SDD available)

By calibrating the LWR parameters (essentially w and Qmax since V0 has little influence),
one obtains an unbiased estimate for ρmax
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.5 Properties of the Triangular FD

Downstream shock front seen microscopically

I Infinite aceleration (softens for nontriangular FD)

I The upstream front is always sharp irrespective of the FD and corresponds to an
infinite deceleration
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.6 Bottlenecks – an overview

6.6 Bottlenecks – an overview

Types of bottlenecks:
I Lane and flow-conservative bottlenecks: no source terms, bottleneck effect only by
spatial change of parameters
I Lane closure bottlenecks: bottleneck effect by source terms in the effective LWR
while the LWR for the total quantities has no sources (flow-conservative)
I Ramp bottlenecks: bottleneck effect by source terms
I Temporary bottlenecks such as traffic lights.
Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.6 Bottlenecks – an overview

Classic flow-conserving bottleneck

Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.6 Bottlenecks – an overview

Classic flow-conserving bottleneck

I The bottleneck with capacity

Cbottl = LQmax has a different FD with
a lower capacity than the other sections
I Definition of bottleneck: locally reduced
capacity
I If congestion arises, the bottleneck emits
information in both directions (why?).
⇒ QIb and QIII are equal to the
bottleneck capacity QII
Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.6 Bottlenecks – an overview

Lane-closing bottleneck
Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.6 Bottlenecks – an overview

Lane-closing bottleneck

I Same FD per lane but different FD for

total quantities

I Bottleneck capacity: Cbottl = L2 Qmax

I Maximum-flow state in the whole

downstream region, in contrast to the
classic bottleneck
Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.6 Bottlenecks – an overview

On-ramp bottleneck
Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.6 Bottlenecks – an overview

On-ramp bottleneck

I All roAd sections have the same FD but

there is a source term in the continuity
equation

I Bottleneck capacity LQmax − Qrmp

I Continuous transition
congested-maximum flow state in
merging region
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.7 Traffic Lights

6.7 Traffic Lights: Temporary bottlenecks

I In the LWR approach, a traffic light is a temporry bottleneck with capacity zero
I When the light becomes green, the stationary downstream jam front congested →
free starts to move upstream at velocity w
I Generally, downstream fronts are either “pinned” at the bottleneck or move upstream
at velocity w
? There is a single situation where a downstream front may move downstream. Which?
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.7 Traffic Lights

Calculation of the total waiting time in one red phase

I ρmax times the vertical extension of the red triangle (e.g., from E to D) gives the actual
number of stopped vehicles per lane
I The area of the triangle gives the total waiting time per lane
I This area can be easily calculated by adding the areas of the two rectangular triangles ODE
and DEF’
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.7 Traffic Lights

Several traffic lights

Draw a triangular FD with the traffic states 1 to 4 and also denote graphically the
different shock-wave propagation velocities
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.8 Examples: 1.accident

6.8 Examples: 1.accident

I Single-lane obstruction between 15:00 and 15:30 pm

I Constant inflow 3 024 veh/h
I Triangular FD parameters leff = 8 m, T = 1.5 s, and V0 = 28 m/s
1. Does the road capacity prior to the accident satisfy the demand?
! Maximum flow per lane: Qmax = V0 TV+l 0
eff
= 2 016 veh/h.
Capacity before the accident: C = 2Qmax = 4 032 veh/h.
This exceeds the traffic demand 3 024 veh/h, so no jam.
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.8 Examples: 1.accident

2. Show that the accident leads to a traffic breakdown. Calculate the total and effective
flows in all sections.

I Bottleneck capacity Cbottl = Qmax = 2 016 veh/h is smaller than inflow Qin ⇒ traffic
breakdown.
I Upstream free flow controlled by inflow, Qtot
1 = 3 024 veh/h, and congested flow as well as
the flow in all following segments by the bottleneck: Qtot tot
2 = Q3 = Cbottl = 2 016 veh/h

I Per lane, the effective flows are Q1 = 1 512 veh/h, Q2 = Q3 = 1 008 veh/h
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.8 Examples: 1.accident

3. Calculate the propagation velocity of the upstream jam front

I Bedides the flow, we need the densities. They are given by the suitable branch of the inverse
FD:
Q1 1 − Q2 T
ρ1 = = 15 veh/km, ρ2 = ρcong (Q2 ) = = 72.5 ve/km
V0 leff

I Propagation velocity of upstream jam front:

Q2 − Q1
cup = c12 = = −8.77 km/h
ρ2 − ρ1
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.8 Examples: 1.accident

4. Calculate the velocity of the moving downstream front once the obstruction has been
removed and the time for complete dissolution of the jam.

Once the obstruction has been removed, the maximum flow state always arising at the
dowwnstream end of jams is over both lanes, so the new flows downstream of the congestion are
Qtot
4 = C and Q4 = C/2 = Qmax = 2 016 veh/h and, from the free branch,
ρ4 = Q4 /V0 = 20 vehicles/h. Thus,

Q4 − Q2
cdown = c = c23 = = −19.2 km/h
ρ4 − ρ2
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.8 Examples: 1.accident

5. Visualize the spatiotemporal

dynamics of the jam by drawing
its boundaries in a space-time
diagram.
6. How much time does a vehicle
need to traverse the 10 km long
road section if it enters
at 15:30 h?
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.8 Examples: 1.accident

Example 2: Uphill Grade and Lane Drop, Q1

Triangular FD (per lane) as follows:

 
60 km/h Section III 1.9 s III
V0 = , T = , leff = 10 m
120 km/h other sections 1.5 s others

1. Calculate the local road capacity and identify possiblebottlenecks

! capacity is number of lanes times Qmax where Qmax is equal for the Sections I, II, and
IV; bottlenecks are local capacity drops, i.e., beginning of the Sections II and III →
tutorial to this lecture
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.8 Examples: 1.accident

Example 2: Uphill Grade and Lane Drop, Q2

Triangular FD (per lane) as follows:

 
60 km/h Section III 1.9 s III
V0 = , T = , leff = 10 m
120 km/h other sections 1.5 s others

2. At 4:00 pm, the total traffic demand at x = 0 increases abruptly from 2 000 veh/h to
3 600 veh/h. Does this cause a breakdown? If so, at which time and where?
! Answer: Check where, when going from upstream to downstream (why?) the demand
exceeds the capacity for the first time. The relevant bottleneck is said to be activated
→ tutorial to this lecture
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.8 Examples: 1.accident

Example 2: Uphill Grade and Lane Drop, Q3

Triangular FD (per lane) as follows:

 
60 km/h Section III 1.9 s III
V0 = , T = , leff = 10 m
120 km/h other sections 1.5 s others

3. Calculate the dynamics of the developing congestion if the inflow remains constant
! The downstream front is pinned at the activated bottleneck. The upstream front is
determined by the shockwave formula. You need to distinguish the propagation in the
Regions II and I! → tutorial to this lecture
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.9 Numerics of the LWR

6.9 Numerics of the LWR

I: Mathematical background
∂Qe (ρ)
Mathematically, the LWR model equation ∂ρ ∂t + ∂x is a hyperbolic partial differential equation
(PDE) for ρ(x, t) in the special form of a conservation law. This PDE can be solved (the
problem is well posed; the solution exists and is unique as the math people say) provided

I The initial condition ρ(x, 0) at time t = 0 is completely known for all x ∈ [0, Lroad ] along
the road
I In case of free traffic ρ(0, t) < ρc , the upstream boundary condition (BC)
ρ(0, t) = ρfree (Qup (t)) is given by the traffic demand Qup per lane
I In case of a downstream congestion, the downstream BC ρ(Lroad , t) = ρcong (Qdown (t)) is
determined by the maximum flow this congestion can take.

I When solving a conservation law, it is crucial to take into account the direction of
information flow.,
I Depending on the situation, 0, 1, or 2 BC apply
Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.9 Numerics of the LWR

General
LWR models

I Cell length ∆x, update

time interval ∆t,
approximated density
ρkj = ρ(k∆x, j∆t)

I take into account signal

propagation directions
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.9 Numerics of the LWR

General LWR models (ctnd): discretisation in space

I Approximate the spatial derivative ∂Q

∂x by first-order finite differences taking account
the signal propagation (essentially Godunov’s method)
I If information propagates downstream (free traffic), use downwind finite
∂
differences (“with the wind, with the stream”) for ∂x to “catch” this information:

∂Q(k∆x, j∆t) Qk,j − Qk−1,j

≈
∂x ∆x
I If information propagates upstream (congested traffic), use upwind finite
differences (“against the wind”):

∂Q(k∆x, j∆t) Qk+1,j − Qk,j

≈
∂x ∆x
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.9 Numerics of the LWR

General LWR models (ctnd): discretisation in time

I Explicit numerical scheme: only past and present needed

I First-order scheme: errors for integrating a certain time interval decrease linearly
with ∆t and ∆x if both tend to zero
I The Euler method is the simplest of such schemes: f (t + ∆t) ≈ f (t) + f 0 (t)∆t for
any f (t)
⇒


∆t Qk−1,j − Qk,j
free
ρk,j+1 = ρk,j − ∆x ,
Qk+1,j − Qk,j congested
Qk,j+1 = Qe (ρk,j+1 )

Courant-Friedrichs-Lévy (CFL) stability criterion:

∆t ≤ |c|max ∆x = V0 ∆x
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.9 Numerics of the LWR

Numerics III: Supply-Demand-Method for Triangular FDs

General scheme also applies to triangular FDs. However, near capacity, it becomes tricky
to determine whether downwind or upwind finite differences to apply (why?). The
supply-demand method gives a specialized simplified procedure for triangular FDs:
1. Define the supply (maximum flow the downstream cell can receive) and demand
(maximum flow the upstream cell can deliver) functions with capacity Ck = Lk Qmax :

S(k) = min(Ck , Lk Qcong (ρk )),

D(k) = min(Ck , Lk Qfree (ρk ))
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.9 Numerics of the LWR

Supply-Demand-Method for Triangular FDs (ctnd)

2. As in any trading, the actual flow (amount of delivered products) is given by the minimum of
supply and demand. For the two boundaries of cell k:
Qtot,up
k = Qtot,down
k−1 = min (Sk , Dk−1 ) ,
Qtot,down
k = Qtot,up
k+1 = min (Sk+1 , Dk ) .

3. Explicit first-order as in the general case:

1  
ρk (t + ∆t) = ρk (t) + Qtot,up
k − Qtot,down
k ∆t,
Lk ∆xk
Qk (t + ∆t) = Qe (ρk (t + ∆t)) .
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.9 Numerics of the LWR

Cell-transmission model for networks

S3 = min(C3 , L3 Qcong (ρ3 )),

D3 = min(C3 , L3 Qfree (ρ3 )),
D12 = min(C1 , L1 Qfree (ρ1 ) + min(C2 , L2 Qfree (ρ2 ),
tot,up
Q3 = min(S3 , D12 ),
tot,down
Q3 = min(S4 , D3 ),
1  tot,up tot,down

ρtot
3 (t + ∆t) = ρtot
3 + Q3 − Q3 ∆t
∆x3
In case of congestion, only the sum Qtot,up
3 is defined and the supply distributed to the
demands D1 and D2 according to traffic regulations/priority rules.
 Traffic Flow Dynamics 6. The Lighthill-Whitham-Richards (LWR) Model 6.9 Numerics of the LWR

Cell transmission model for other bottlenecks

I Bottlenecks in the stricter sense and also changes in the number of lanes? are
automatically included in the supply-demand model for straight roads
I Merge bottlenecks? are just two-in-one nodes where the priority (in contrast to
regulations) is given to the ramp
I Diverges? such as cell 1 → cells 2, 3 require the diverging fraction λ3 as additional
input. Congestion arises if λ3 D1 > S3 or (1 − λ3 )D1 > S2 . If λ3 D1 > S3 , we have

S3
Qtot,down
1 =
λ3
leading to a spill-back bottleneck

1. Make this formula plausible! On average, a flow S3 /λ3 can pass the diverge such that Link 3 is
at its supply limit S3 . The excess link-3 drivers wait until they can enter thereby obstructing also the
upstream link-2 drivers
2. Discuss lane changes as sources of off-ramp bottlenecks Lane changes disturb the flow
reducing the maximum flow. This cannot be modelled by LWR models unless special provisions are taken.