Generalized Macroscopic Traffic Model With Time Delay
Generalized Macroscopic Traffic Model With Time Delay
Generalized Macroscopic Traffic Model With Time Delay
net/publication/259783412
CITATIONS READS
28 697
1 author:
Dong Ngoduy
Monash University (Australia)
130 PUBLICATIONS 2,571 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Dong Ngoduy on 16 March 2016.
Abstract The effect of delay or reaction time on traffic flow larly, Bando et al. [1] have found that large delay contributes
dynamics has been investigated widely in literature using significantly to the formation of traffic clusters using linear
microscopic traffic models. Recent studies using second or- analysis and numerical simulations of a so-called optimal
der Payne-type models have shown analytically that, on macro- velocity model (OVM). Recently, Treiber et al. have derived
scopic scale, time delay does not contribute to whether traf- a so-called (microscopic) intelligent driver model (IDM) in
fic instabilities occur. This paper will attempt to show that it which the finite delay is taken into account [37]. The authors
all depends on the (macroscopic) model used for the analy- have shown that delay has significant effects on the tran-
sis that delay does have effect on traffic instabilities or not. sitions of congested traffic states. Using microscopic data
To this end, we will formulate a generalized (linear) stability obtained from the helicopter, Hoogendoorn et al. have pro-
condition for a second order macroscopic model with delay vided an empirical evidence of the effects of delay on traf-
and investigate analytically the effect of such delay on traf- fic stability [9], which also supported the results in [1, 37].
fic instabilities in some specific macroscopic models. It is In a similar line, Ngoduy has described analytically the ef-
found that the choice of the equilibrium speed function in a fect of delay in a mixed (heterogeneous) traffic flow using
(second order) macroscopic model will determine how delay a multi-class IDM [20]. General conclusions from the nu-
affects such (linear) stability condition. merical and analytical results using microscopic models in
literature are that delay is often taken the values in order
Keywords Macroscopic models, traffic instabilities, linear
of one second and increasing delay will destabilize traffic
analysis, delay
flow. In contrast to such findings from microscopic model
based analysis, Tampere et al. [29] and Ngoduy [22] have
1 Introduction found that delay does not contribute to whether (linear) traf-
fic instabilities occur. Nevertheless, the linear stability result
From the machine point of view, delay occurs due to the time does not necessarily means that delay has no effect at all
needed for devices to react to the front stimulus whereas on stability since the validity of the linear stability result in
from the human operation point of view, such delay is de- [22, 29] is only limited in a specific model which has been
fined as a physiological parameter reflecting the period of derived from a specific car-following model where the inter-
time between perceiving a stimulus and performing a rele- action is only dependent on the space headway distribution
vant action. The effect of delay on traffic flow dynamics has (e.g. OVM). Including the speed difference in the interaction
been investigated both analytically and numerically in traf- could lead to a strong dependence of stability on delay. This
fic flow literature using microscopic traffic models. Particu- is the topic of this paper.
To contribute to the state-of-the-art in traffic flow mod-
We wish to thank the financial support from the UK research council
EPSRC, grant EP/J002186/1 eling, this paper will attempt to formulate the linear stabil-
∗
ity analysis for a general class of second order macroscopic
Corresponding author
Institute for Transport Studies
models with delay. We will show in Section 3 that the choice
University of Leeds of the equilibrium speed function will determine weather or
E-mail: [email protected] not delay affects the (linear) traffic instabilities. Section 4
describes some simulation results that show the effects of
2 D. Ngoduy∗
∂r
delay on the evolution of traffic instabilities. Finally, Sec- local density and speed gradients rx = ∂x and Vx = ∂V ∂x ,
tion 5 concludes the paper and discusses the future research. respectively, and the nonlocal traffic density ra and nonlocal
speed Va , which are determined at a location xa in front of
the actual position x: xa = x + d (d > 0).
2 Model equations
– Payne-type model:
In general, there are two types of traffic flow models which ′
can be used for the stability analysis: microscopic models Ve (r) − V V (r) ∂r
A (r, V, ra , Va , rx , Vx ) = + e
and macroscopic models. Microscopic models describe traf- τ 2rτ ∂x
fic flow at high level of detail such as the movement of in- – Nonlocal model:
dividual vehicles [2, 4, 10–12, 33, 37]. Macroscopic mod-
Ve∗ (r, V, ra , Va ) − V 1 ∂P
els represent traffic flow at low level of detail by aggre- A (r, V, ra , Va , rx , Vx ) = −
τ r ∂x
gate traffic variables such as flow, mean speed and density
[3, 5, 6, 8, 15, 17, 23, 24, 26, 28, 30–32, 34, 39–41]. Both where Ve (r) and Ve∗ (.) denotes the equilibrium speed func-
linear and nonlinear stability analysis methods have been ap- tion in either model type and τ denotes the relaxation time.
plied to either microscopic models [10, 13, 14, 20, 27, 35– P denotes the so-called traffic pressure, which is defined as
38, 42] or macroscopic models to study traffic instabilities P = rΘ(r) with Θ being the density dependent speed vari-
′ .
[7, 16, 18, 19, 21, 23, 24, 29, 36, 38]. However, it is also ance. Ve (r) = dV
dr .
e
worth mentioning that stability analysis using microscopic The delay is defined as below:
models usually refers to the string stability of a platoon of – For human-driven vehicles: The delay occurs due to the
vehicles following each other, which is also equivalent to physiological (human) perception which reflect the pe-
the flow stability analysis using macroscopic models. For a riod of time between perceiving a stimulus and perform-
detailed overview of the different model classes as well as ing a relevant action (e.g. acceleration/deceleration due
general stability analysis of each model class, we refer to to traffic situations downstream).
[36]. – For intelligent vehicles: The delay occurs due to the time
In principle, macroscopic models are further classified needed for sensing, computation and actuation in com-
into two classes. The first order class of models is based on puterized devices.
a kinematic wave theory, which is originally developed by
Lighthill and Whitham [15] and independently by Richards The delay is included in the model equation (2) as:
[28], here after LWR for short. In the LWR model, the local ∂V ∂V
seed and flow are assumed a empirical (equilibrium) func- +V = A (r, V, ra , Va , rx , Vx ) |t−T , (3)
∂t ∂x
tion of the density. This assumption implies that the speed where T denotes [Tr , Tv ], in which Tr and Tv are delay with
or flow adapts instantaneously to the density changes so the respect to the density and speed, respectively. By definition
model is not able to replicate traffic instabilities. The sec- in Orosz et al. [25], Tr = Tv > 0 for intelligent vehicles
ond order class of traffic models has been developed in an whereas Tr > 0 and Tv = 0 for human driven vehicles
attempt to replicate such instabilities due to perturbations as- because drivers are aware of their own speed immediately.
sociated with acceleration/deceleration, lane-changes or dis- First order Taylor expansion of equation (3) leads to:
turbances of traffic at an on-and off-ramp bottleneck (where
stop-and-go waves and spontaneous formation of traffic jams ∂V ∂V
+V = A (r, V, ra , Va , rx , Vx )
are observed). Since our aim is to study the effect of delay ∂t ( ∂x )( )
∂A ∂A ∂r ∂r
on traffic instabilities, this paper will focus on mathematical −Tr + +V
equations describing the dynamics of traffic low in second ∂r ∂ra ∂t ∂x
( )( )
order macroscopic models with delay. ∂A ∂A ∂V ∂V
−Tv + +V (4)
Let r = r(x, t) and V = V (x, t) denotes, respectively, ∂V ∂Va ∂t ∂x
the density and the mean speed at location x and time instant To substitute equation (1) to equation (4) to obtain a generic
t. The macroscopic governing equations for the dynamics of speed dynamic equation with delay:
r and V can be written in the form of a system of partial
differential equations below: ∂V ∂V
+V = Ã (r, V, ra , Va , rx , Vx ) (5)
∂r ∂(rV ) ∂t ∂x
+ =0 (1)
∂t ∂x where
∂V ∂V ( )
+V = A (r, V, ra , Va , rx , Vx ) , (2) Tr r ∂A
+ ∂A
∂t ∂x A ∂r ∂ra ∂V
à = ( )+ ( ) ,
where A(.) denotes the acceleration function, which is de- ∂A ∂A ∂A ∂A ∂x
1 + Tv ∂V + ∂Va 1 + Tv ∂V + ∂Va
pendent on the local traffic density r and local speed V , the
Generalized macroscopic traffic model with time delay 3
v r ∂r V0 τ
200 Tv=0,Tr=1.5sec
Stable which returns to the original linear stability condition of
150
the nonlocal model without delay.
Stable
100
Unstable
50
4 Numerical studies
0
10 20 30 40 50
Density (veh/km) In this section, we will numerically illustrate the perfor-
mance of our nonlocal Treiber-type model with delay in
Fig. 1 Linear stability diagram order to support our analytical results found above. To
this end, let us simulate the proposed model with model
parameters given above and delay Tr = 1.0sec. Accord-
Condition (9) becomes: ing to our analytical results (i.e. Figure 1), traffic is un-
( ) [
′ 2 Tr
(
Vmax − V0
)] stable when the initial density is between [11, 40]veh/km
r0 Ve 1+ 1+2 ≤ for Tr = 0sec and [11, 45]veh/km for Tr = 1.0sec.
τ V0
[ ′
] The initial density is chosen as:
∂P 2d (Vmax − V0 ) r0 Ve rjam V0
− √ − (12)
∂r V0 τ πα rjam − r0 r(x, 0) = r0 + δr
[ ( ( )) ( ( ))]
−2 160 5L −2 40 11L
It is clear that the delay might affect the linear stability cosh x− − 0.25cosh x− ,
L 16 L 32
condition in the nonlocal Treiber-type model, depending
on the values of the homogeneous stationary equilibrium where L is the length of the simulated freeway (L =
speed V0 . In the ensuing paper, let us investigate the ef- 18km) and δr = 10veh/km . Let us consider the follow-
fect of delay on traffic instabilities using the nonlocal ing various initial homogeneous and stationary equilib-
Treiber-type model. rium conditions:
For stationary and homogeneous traffic density, the equi- – Free-flow stable condition: r0 = 15veh/km. In this
librium speed function of the nonlocal model is defined situation, traffic is linearly stable according to our
as: analytical condition. Figure 2 definitely shows that
( √ ) traffic is stable in for both situations, with and with-
Ṽ02 2
4Vmax out delay.
Ve (r) = −1 + 1 + , (13)
2Vmax Ṽ02 – Unstable condition: r0 = 30veh/km. According to
Figure 1, this density belongs to the linearly unstable
where regime for both Tr = 0sec and Tr = 1.0sec. Figure
3 shows that traffic is linearly unstable in both sit-
( )√
1 1 1 α(rjam ) uations, with and without delay. More particularly,
Ṽ0 = − , an initial small perturbation has led to stop-and-go
h r rjam α(r)
waves. However, the delay results in larger ampli-
The model parameters used in the rest of this paper are: tude of the stop-and-go waves, measured by the peaks
– Free-flow speed: Vmax = 100km/h. of the density oscillations as shown in Figure 4.
– Jam density: rjam = 140veh/km – Congested stable condition for Tr = 0 and unstable
– Critical density: rcr = 35veh/km condition for Tr = 1.0sec: r0 = 41veh/km. Figure
– Safe-time headway: h = 1.4sec 5 indicates that for this initial density value, traffic is
– Relaxation time: τ = 25sec linearly stable without delay but is linearly unstable
– Delay time: Tr varies between [0, 1.5]sec with delay as the delay has expanded the unstable
Figure 1 describes the linear stability diagram for traffic regime (which conforms to Figure 3).
flow with delay using the nonlocal Treiber-type model. It – Congested stable condition for Tr = 1.0sec: r0 =
can be seen that the (linear) unstable traffic regime is ex- 50veh/km. As this density belongs to the linearly sta-
panded with the increased delay. It is worth mentioning ble regime of our model with Tr = 1.0sec, the per-
that in the linear (free-flow) stable regime, the delay has turbation will die out in time and traffic is linearly
no impact since in this regime V0 ≈ Vmax so condition stable, which is clearly shown in Figure 6.
Generalized macroscopic traffic model with time delay 5
Density (veh./km)
80 Tr=0sec
60
40
20
0
0 50 100 150
Space (100m)
Tr = 0sec
Fig. 4 Time profile of traffic density, at t = 300sec
Tr = 1.0sec
Fig. 2 Traffic is stable in free-flow condition Tr = 0sec
Tr = 1.0sec
Tr = 0sec
Fig. 5 Traffic is stable without delay and unstable with delay
Tr = 1.0sec
Fig. 3 Instabilities occur in the immediate condition (meta-stable) Tr = 1.0sec
cruise control vehicles. Transportmetrica 8, 43–60 35. Treiber, M., Kesting, A.: Evidence of Convective
(2012b) Instability in Congested Traffic Flow: A System-
20. Ngoduy, D.: Analytical studies on the instabilities atic Empirical and Theoretical Investigation. Trans-
of heterogeneous intelligent traffic flow. Communi- portation Research Part B 45, 1362–1377 (2011)
cations in Nonlinear Science and Numerical Simu- 36. Treiber, M., Kesting, A.: Traffic Flow Dynamics.
lation 18, 2699–2706 (2013a) Springer, Germany (2013)
21. Ngoduy, D., Hoogendoorn, S.P., Liu, R.: Continuum 37. Treiber, M., Kesting, A., Helbing, D.: Delays, in-
traffic flow modelling of cooperative traffic systems. accuracies and anticipation in microscopic traffic
Physica A 388, 2705–2716 (2009) model. Physica A 360, 71–88 (2005)
22. Ngoduy, D., Tampere, C.M.J.: Macroscopic effects 38. Wilson, R.E., Berg, P., Hooper, S., Lunt, G.: Many
of reaction time on traffic flow characteristics. Phys- neighbour interaction and non-locality in traffic
ica Scripta 80, 025802–025809 (2009) models. The European Physical Journal B 39, 397–
23. Ngoduy, D., Tampere, C.M.J.: Macroscopic effects 408 (2004)
of reaction time on traffic flow characteristics. Phys- 39. Wong, G.C.K., Wong, S.C.: A Multiclass Traffic
ica Scripta 80, 025,802–025,809 (2009) Flow Model-an Extension of LWR Model with Het-
24. Ngoduy, D., Wilson, R.E.: Multi-anticipative non- erogeneous Drivers. Transportation Research Part
local second order traffic model. Computer-Aided A 36, 763–848 (2002)
Civil and Infrastructure Engineering in press (2013) 40. Zhang, P., Wong, S.: Essence of conservation forms
25. Orosz, G., Wilson, R.E., Stephan, G.: Traffic jams: in the travelling waves solutions of higher order
dynamics and control. Philosophical Transactions traffic models. Physical Review E 74, 0261,091–
of the Royal Society A 368, 4455–4479 (2010) 02610,910 (2006)
26. Payne, H.J.: Models for Freeway Traffic Control. 41. Zhang, P., Wong, S.C., Dai, S.Q.: A conserved
Mathematical Models of Public Systems 1, 51–61 higher-order anisotropic traffic flow model: De-
(1971) scription of equilibrium and non-equilibrium flows.
27. Peng, G., Nie, F., Cao, B., Liu, C.: A drivers mem- Transportation Research Part B 43, 562–574 (2011)
ory lattice model of traffic flow and its numerical 42. Zheng, L.J., Tian, C., Sun, D.H., Liu, W.N.: A new
simulation. Nonlinear Dynamics 67, 1811–1815 car-following model with consideration of antici-
(2012) pation driving behavior. Nonlinear Dynamics 70,
28. Richards, P.I.: Shock Waves on the Highway. Oper- 1205–1211 (2012)
ations Research 4, 42–51 (1956)
29. Tampere, C.M.J., Hoogendoorn, S.P., van Arem, B.:
A behavioural approach to instability, stop and go
waves, wide jams and capacity drop. In: Proceed-
ings of the 16th International Symposium on Trans-
portation and Traffic Theory , pp. 205–228 (2005)
30. Tang, T.Q., Huang, H.J., Xu, G.: A new macro
model with consideration of the traffic interruption
probability. Physica A 387, 6845–6856 (2008)
31. Tang, T.Q., Li, C., Huang, H., Shang, H.: A new fun-
damental diagram theory with the individual differ-
ence of the driver’s perception ability. Nonlinear
Dynamics 67, 2255–2265 (2012)
32. Tang, T.Q., Li, C.Y., Wu, Y., Huang, H.J.: Impact of
the honk effect on the stability of traffic flow. Phys-
ica A 390, 3362–3368 (2011)
33. Tang, T.Q., Wang, Y., Yang, X., Wu, Y.: A new car-
following model accounting for varying road condi-
tion. Nonlinear Dynamics 70, 1397–1405 (2012)
34. Treiber, M., Hennecke, A., Helbing, D.: Derivation,
Properties and Simulation of a Gas-kinetic-based,
Non-local Traffic Model. Physical Review E 59,
239–253 (1999)