![](bg2.png)
Figure 1: Conceptualized distributed transportation network.
significantly higher than in the user equilibrium
[Roughgarden and Tardos 2000]. The main
drawback of this approach is that after a while a user
may be unsatisfied and stop using the route guidance
system.
On the other hand, when minimizing individual
journey times of each driver, we can be far from the
system equilibrium. Many algorithms try to
minimize individual journey times, without taking
into account the effects of their route
recommendation. Under current systems, many
drivers could be given exactly the same route
recommendation. Therefore, assuming similar mean
speed, some drivers may always stay together on
their trip (platooning), and this may possibly lead to
congestion. While Adler and Blue [1998] call this
phenomenon oversaturation, Ben-Alkiva. De Palma
and Kaysi [1991] call it overreaction, since too
many drivers follow the recommended route. To
deal with this problem, it is essential to split
platoons over several paths (multiple path routing).
Many authors have proposed different approaches to
deal with multiple path routing. Rilett and Van
Aerde [1996] suggest adding individual random
error terms to the road travel times broadcast by a
central controller, in order to cause the in-vehicle
computers to choose different paths. Lee [1994]
computes k-shortest every ten minutes and then
distributes drivers over them every two minutes,
considering the current travel times on these paths.
Another popular approach is to route drivers
along the paths of so-called user equilibrium, so that
no driver can get a quicker path through the network
by unilaterally changing his route [Fresz T. L.,
1985]. This concept was introduced to model natural
driver behaviour and it has been studied extensively
in the literature.
In our work we assume that connected users are
provided with an Intelligent Traveller Information
System (ITIS) capable of providing route guidance
and/or traffic advice both pre-trip and while en-
route. ITIS is a term coined by Adler and Blue
[2002] to describe next generation information
devices that can gather and process information as
well as learn and represent user preferences and
behaviour.
3 PROBLEM DEFINITON AND
PROPERTIES
A traffic network, represented by a direct graph G=
(N,A), consists of a set of |N| of nodes and a set of |A|
links. Consider a situation in which a vehicle with
ITIS (Intelligent Traveller Information System, see
Figure 1) is currently travelling on link (i,s) towards
the destination node d, we want find on which link
(s,j) the vehicle should enter next, so as to minimize
the expected travel time to the destination node d.
It is assumed that the local controller (i.e. a
vehicle with ITIS) has available complete
information on the topology of sub network G′ (the
network representing all candidate paths from origin
o to destination d of the interested vehicle) and
current estimates of travel times on individual links.
Traffic flows have two important features that
make them difficult to study mathematically. One is
“congestion”, and the other is “time”. Congestion
captures the fact that travel times increase with the
amount of flow on the streets, while time refers to
the movement of vehicles along paths as “flow over
time”.
Congestion implies that transit time t
e
is not a
constant, but monotonically increases with the
augmentation of the flow value x
e
..
Flow variation over time is an important feature
in network flow problem arising in various
applications such as road or air traffic control,
production systems, and communication network.
A DECENTRALIZED ROUTE GUIDANCE ALGORITHM IN URBAN TRANSPORTATION NETWORKS
319