\ .
|
|
(4)
For 2 | = , the solution is GUE,
2
GUE 2
32 4
( ) exp
2
P s s s
|
=
\ .
|
|
(5)
For 4 | = , the solution is GSE,
18
4
GSE 6 3
2 6
( ) exp
3 9
2
4
P s s
|
=
\ .
s
|
|
(6)
As discussed by Rawal and Rodgers
[7]
, the move-
ments of different size vehicles can be abstracted into
point particles, because only the spacing distribution is
interested here. This analysis assumes that the queuing
dynamics of vehicles at a signalized junction can also
be approximately modeled by this Coulomb gas model.
The single-particle term in Eq. (1) can be viewed as
reflecting the tendency to drive closer while the repul-
sive two-body term in Eq. (1) indicates the tendency to
maintain a safe distance. The basic instinct of a driver
is to maintain a small but safe gap between his vehicle
and the leading vehicle, especially when queuing. The
superposition of these two potentials, which creates an
overall repulsion for small spacings and an attraction
for large ones, expresses the observations that very
small or very large spacings between queuing vehicles
are unlikely. However, no driver can always keep an
ideal headway due to disturbances (unexpected accel-
eration/deceleration of the leading vehicle, occasional
wanderings of the mind, etc.). Thus, the vehicles
(particles) are also simultaneously perturbed by
environment.
This analogue shows that the spacing distribution of
parked vehicles is in agreement with the GUE distribu-
tion
[8]
, when the empirical spacing distribution of
queuing vehicles systems is assumed to fit one of the
Wigner surmises in Eqs. (3)-(6).
Tsinghua Science and Technology, April 2009, 14(2): 252-254 254
2 Comparison with Empirical
Results
To test this conjecture, 700 sample spacings of queuing
vehicles were collected at several different signalized
junctions in Beijing, China. Details of the data collec-
tion were given earlier
[10]
.
The average observed spacing was 1.43 m. Figure 2
shows the probability distribution function, P(s), in the
form of a normalized histogram where s is defined as
the ratio of the spacing to the mean value.
Fig. 2 Normalized spacing distribution of queuing
vehicles at signalized junctions compared with Poisson,
GOE, GUE, and GSE
In the Coulomb gas model, the inverse temperature
| of the gas characterizes the degree of repulsion.
Figure 2 shows
Poisson
( ) P s ,
GOE
( ) P s ,
GUE
( ) P s , and
GSE
( ) P s as well as the empirical spacing distribution
histogram. Unlike the parked vehicle data, the vehi-
cle-queuing data more closely fits the GSE model
(passes the Kolomogorov-Smirnov test with a default
level of 0.05) instead of the GUE model, even though
in these two models, the drivers all aim to drive close
but not too close. This suggests that the vehicle queu-
ing process at signalized junctions also experiences
RMT-like fluctuations.
The spacing distributions of parked vehicles and
queuing vehicles most likely differ due to differences
in the driving patterns for these two scenarios. When
parking, drivers would like to try several times and
move back-and-forth to adjust the gaps so as to park in
an ideal position; while in vehicle-queuing scenarios,
drivers cannot back up to adjust the spacing. Thus,
when parking, the repelling force from neighboring
vehicles is relatively loose; while in queues, the re-
pelling force from neighboring vehicles is rather
rigid. Talbot et al.
[5]
pointed out that such a differ-
ence can be modeled by different inverse temperatures
| . At low temperatures ( | is larger), the charges
tend to be regularly spaced in a crystalline lattice ar-
rangement and the randomness of the positions of the
charges is small. At higher temperatures ( | is
smaller), the charges fluctuations become more intense.
Thus, for parking scenarios, the GUE model with
2 | = fits best with the empirical data while for vehi-
cle queuing, the GSE model with 4 | = fits best with
the empirical data.
In different cities, the spacing within vehicle queues
may still fit GSE distribution but with different means.
Further measurements will test this hypothesis by ve-
hicles queuing data collected in other cities.
References
[1] Schonhof M, Helbing D. Empirical features of congested
traffic states and their implications for traffic modeling.
Transportation Science, 2007, 41(2): 135-166.
[2] Helbing D, Treiber M, Kesting A. Understanding interarri-
val and interdeparture time statistics from interactions in
queuing systems. Physica A, 2006, 36(1): 62-72.
[3] Zhang G H, Wang Y H, Wei H, et al. Examining headway
distribution models using urban freeway loop event data.
In: Transportation Research Board Annual Meeting. 2007.
[4] Renyi A. On a one-dimensional problem concerning ran-
dom space filling. Selected Translations in Mathematical
Statistics and Probability, 1963, 4: 203-218.
[5] Talbot J, Tarjus G, Van Tassel P R, Viot P. From car parking
to protein adsorption: An overview of sequential adsorp-
tion processes. Colloids and Surfaces A, 2000, 165(1-3):
287-324.
[6] Lee J W. Reversible random sequential adsorption on a
one-dimensional lattice. Physica A, 2004, 331(3): 531-537.
[7] Rawal S, Rodgers G J. Modelling the gap size distribution
of parked cars. Physica A, 2005, 346(3-4): 621-630.
[8] Abul-Magd A Y. Modelling gap-size distribution of parked
cars using random-matrix theory. Physica A, 2006, 368(2):
536-540.
[9] Mehta M L. Random Matrices, 3rd edition. Boston, USA:
Academic Press, 2004.
[10] Su Y, Wei Z, Cheng S, et al. Departure headways of mixed
traffic flow at signalized intersections: Distributions,
simulations and validations. In: Transportation Research
Board Annual Meeting CD. 2008.