VOL. 2, NO.

9, September 2011 ISSN 2079-8407

Jour nal of Emer gi ng Tr ends i n Comput i ng and I nf or mat i on Sci ences

2009-2011 CIS Journal. All rights reserved.

http://www.cisjournal.org
Rough Set adaptive in the Model Based of Cellular Automata
and Multi-Agents

Yasser F. Hassan
Department of Mathematics & Computer Science,
Faculty of Science, Alexandria University, Egypt,
[email protected]

ABSTRACT
The need for intelligent systems has grown in the past decade because of the increasing demand on humans and
machines for better performance. The researchers of AI have responded to these needs with the development of
intelligent hybrid systems. This paper describes the modeling language for interacting hybrid systems in which we
will build a new hybrid model of cellular automata, multiagent technology and rough set theory. Therefore, in our
approach, cellular automata form a useful framework for the muliagent simulation model response it in simulated cars
in traffic system which lies in adapting the local behavior of individual agent using rough sets to provide an
appropriate system-level behavior in grid of interacting organisms. The modeled development process in this paper
involves simulated processes of evolution, learning and self-organization. The main value of the model is that it
provides an illustration of how simple learning processes may lead to the formation of the state machine behavior,
which can give an emergent to the model.

Keywords: Cellular Automata, Rough Set, Multiagents, Emergence, Traffic System.

1. INTRODUCTION

AI has made great strides in computational
problem solving using explicitly represented knowledge
extracted from the task. If we continue to use explicitly
represented knowledge exclusively for computational
problem solving, we may never computationally
accomplish a level of problem solving performance
equal to humans. From this idea, the paper describes the
development of rough set theory that can be used to
support the assessment of design performance in the
cellular automata model using multiagents technique.
Agents represent objects or people with their own
behavior, think using rough set model and take the
structure of cellular automata lattice.
A Cellular Automaton (CA) [3, 4], as the term
is used in this paper, is a discrete state system consisting
of a countable network of identical cells that interact
with their neighbors. This network can take on any
number of dimensions, starting from a one-dimensional
string of cells. The cellular automata model is perhaps
the simplest, general model available. It is simple in that
the basic units are small, local, finite state machines
(cells). It is general since: cellular automata model is
support universal computation, and the rule represents a
general form of local interaction. The two-dimensional
cellular automata model is a grid of squares, each square
having surrounded adjacent neighbors. A cell occupying
a square is born, lives, or dies based on the number of
living neighbors it has.
In next section, we start with the general model.
This section has subsections: in first and second
subsections, an introduction to the basic notions and
language used in fields that contributed to this study,
namely, rough set theory and cellular automata model
are presented. Followed that, we present a general view
of the combination system of rough set theory and
cellular automata model, in section 3. We present the
model characteristics that display several behaviors
including: reproduction, growth, and mobility in section
4 and some emergent behavior of the new model will be
mentioned. The application of our model to traffic
system design is presented in section 5 with a discussion
of our results. The section has three subsections; first
subsection gives the definition of the traffic flow model
and road network. The following subsection describes
the behavior of at crosscut road and how we can use
rough set theory in the decision-making at road crosscut.
The method proposed undergone experimental
verification and results of those experiments are
mentioned in third subsection. The paper will be
concluded in section 6.

2. Formal Definitions of Proposed Model

In this section, we describe the overall design of
our automaton. A system with a collection of
communicating agents is constructed in this section. We
will consider two-dimensional cellular automata model
which consists of array of cells x(i, j). The use of a cell
as intelligent agent provides an even greater amount of
flexibility to the ability and configuration of the system
itself. Each agent contains the same or different rule
(knowledge) according to which agent states are updated
in a synchronous and local manner. The neighborhoods
of the agent x(i, j) are the von-Neumann neighborhood
of radius r [6], i.e.
} : ) , ( { ) , ( r j n i m n m x j i N
r
s + = (1)
where is the neighborhood relation function of
distance r. We assign r = 1, i.e. the neighbors of an
agent X are the four orthogonal adjacent agents plus the
agent X itself. The agent at cell is called grid
agent if and only if all neighborhood agents
r
N
) , ( j i x

440
VOL. 2, NO. 9, September 2011 ISSN 2079-8407
Jour nal of Emer gi ng Tr ends i n Comput i ng and I nf or mat i on Sci ences

2009-2011 CIS Journal. All rights reserved.

http://www.cisjournal.org

441
) , ( ) , ( j i N n m x
r
e
S f :
exist. Otherwise is called a
boundary agent. Assume that any agent in the lattice is
labeled by its position x = (i, j) where i and j are the row
and column indices. At any given instant, the agent is
assumed to be in one state of the set of states S. Then
abstractly, each agent state function is
) , ( j i x
S A (2)
where A = {a
1
, a
2
,} is a set of actions. The intuition is
that an agent decides what action to perform based on its
state and its environment states. I.e. the agent takes its
current state and an action and maps them to a set of
states S; those that are called result from performing
action a e A in state seS.
The set of agents in the model needs to interact
so that the agents use communication protocol to process
their interactions. The communication is held between
each agent and a set of neighbor agents. Communication
protocols are typically specified at several levels [1]. The
lowest level of the protocol specifies the method of
interconnection; the middle level specifies the format, or
syntax, of the information being transferred; the top level
specifies the meaning of the information.
Once the automaton was embedded in the grid, the
agent began to follow the rule that stored in it. A single
agent cannot do much without interacting with other
agents, and it has no concept of the whole. Yet, in
combination it can play its part in producing complex
results as emergence from local interactions. Briefly,
each agent from the grid able to:
1. exercise a degree of freedom in its operations. It
takes initiative and exercises a non-trivial
degree of control over its own actions.
2. collaborate and exchange information with
other agents in the grid (environment) to assist
other agents in improving their quality of
decision making as well as its own.
3. learn its optimal action by evolving its local
rule.
4. change its state and the states of its immediate
neighbors.
5. copy its rule into a neighboring died agent
(travail self-reproducing). Contention occurs if
more than one neighbor attempts to copy itself
into the same agent. Such a situation is resolved
randomly, i.e. one of the neighbors wins and
copies its rule into the cell.
6. neither read nor write directly to other agents in
the grid except its immediate neighbors.

3. Rough Sets in the Proposed Model

By embedding the method of rough set decision
model within ongoing local control of cellular automata
agent and by allowing of agent/environment interaction
to take place within the system, we can obtain adaptation
of agent behavior, which is reliably good [5, 8, 10].
In our approach, the decision tables define the behavior
of the controller in each agent state and are responsible
for state changes. Each agent has an associated decision
table that defines the combinational actions in this state.
In general cases, the agent followed the cellular
automata rules for interaction with the neighbor agents,
and when it needs to take decision at some spatial
situations, it followed the rules that are induced from the
decision table.
It is necessary for a specification method based
on decision tables to be integrated with agent
descriptions [6]. The minimal decision algorithm
mentioned above can be described easily by rules such
as IF conditional part THEN conclusive part
(decision), and used by each agent. In our approach, the
decision tables define the behavior of the agent in each
state and are responsible for state changes.
Consequently, each agent has an associated decision
table that defines the combinational actions in the
behavior of this agent. This leads to dynamic knowledge
bases, in which the active sets of rules are determined by
the state of the agent. The decision table of the agent
takes the form , where U is the set of
success time steps, A is the condition attributes of the
agent, and d is the decision attribute which describes the
behavior of the agent.
}) { , ( d A U
In our implementation, the rough agent consists of the
following subsystems:



Figure 1: integrated rough set into cellular automata
model.

This method can be considered as semi-
unsupervised algorithm. The basic procedure of an
unsupervised learning method involves grouping inputs
together according to similarity or indicernibility. This
process discovers cases in which a set of input values
typically co-occur and thus tends to nth-order
associations between variable values (where n is the
number of values making up a complete input). The
method can be visualized as an information system
building operation which alternates between a statistical
exploitation phase in which new competitive objects are
produced and a relational exploitation phase in which
new variables are produced. This learning method is
fully incremental. Objects and rules are added to the
information of each agent until the end of time iterations.

VOL. 2, NO. 9, September 2011 ISSN 2079-8407
Jour nal of Emer gi ng Tr ends i n Comput i ng and I nf or mat i on Sci ences

2009-2011 CIS Journal. All rights reserved.

http://www.cisjournal.org

4. Agents Communication

A closer look on emergence proves that
communication is the main reason for achieving global
structure by local interaction, dynamic organization by
simple (communication) rules, and intelligent behavior
of the agent system by dealing with communication
protocols. Thus, as communication is one of the main
methodologies for achieving emergent behavior, an
adequate selection and configuration of communication
protocols is required [9].
A general and conceptual description of the
communication (agent communication protocol) has to
integrate mutual knowledge about its topics (domain),
the general process of the dialog, and its current state.
Thus, the structure of communication protocols may be
divided into a domain dependent (content, problem, and
topic) and a domain independent part consisting of
process knowledge (methods of communication, address
of partners, dialog structure). We assume that an agent
has knowledge like transition rule and a mechanism for
operating on or drawing inferences from its knowledge
using rough sets.
Figure 2 presents two agents communication
technique. A lifeline is any combination of instances.
First, the agent send initial message to start the
communication with another agent. The receiver answers
the initial message by sending message depending on the
understanding of the communication language. If the
receiver understand the language, so it sends proposed
message to ask about the neighborhood relation. The
sender will use neighbors relation function to accept or
reject the proposed message. If the sender accepts the
proposed, the receiver answers by sending its knowledge
using inform message.




Figure 2: The protocol framework of the agents.

5. APPLICATION TO TRAFFIC
SYSTEM CONSTRUCTION

442

In modern societies, the demand for
mobility is increasing daily. Hence, one challenge to
researchers dealing with traffic and transportation is
to find efficient ways to model and predict traffic
flow, even if the behavior of people in traffic is not
a trivial problem. Thus, the modeling and prediction
of traffic flow is one of sciences future challenges.
To be effective, such models have to make
assumptions about the travel demand, and hence
about travel choices and traffic behavior [2, 7].
The present work constructs a traffic model based
on the hybrid system of cellular automata and
multiagent system. Our new model of traffic system
deals with the basic question of decision-making
under such amount of inconsistent information. We
will focus on the decision made by an individual
driver as well as the consideration of the interaction
caused by such a decision on the system as a whole.
We start by proposing a road traffic model suitable
for an urban environment. North, east, south and
west car displacements and road crossings are
VOL. 2, NO. 9, September 2011 ISSN 2079-8407
Jour nal of Emer gi ng Tr ends i n Comput i ng and I nf or mat i on Sci ences

2009-2011 CIS Journal. All rights reserved.

http://www.cisjournal.org
possible. k-states multi-speed (k finite called
maximum velocity v
max
) car motion is found to be a
crucial ingredient to describe highway traffic and
phenomena. The basic idea is to consider the grid of
cells as two types: first type forms a group to
represent buildings (type agent-A) and second agent
type constructs groups of cars representing (type
agent-B). Each site of agent-B type can be in one of
the v
max
states: it may be empty (state = zero or die),
or it may be having an integer velocity between one
and v
max
. This integer number of the velocity is the
number of sites each vehicle advances during one
iteration.
We will use which is called relative speed. A
relative speed measure is to pit the absolute agent
(car) speed against each other in a single elimination
tournament. Let the relative speed of agent or car in
position y to the agent x be:
) ( ) ( ) , ( x v y v y x w = (3)

It is convenient to apply the motion rule to the
lattice with its boundary conditions. We can
consider the velocity v of the agent x is changed
according to the rule:
o +
+ + =
+
) , ( ) , ( ) ( ) (
1
y x w B y x Dist A x v x v
t t t
(4)
where v
t+1
(x) is the velocity of agent x at time step
(t+1), v
t
(x) is the velocity of agent x at time step t, y
is the agent to which the car will move to, w
t
(x, y) is
the relative speed of the car y to the car at cell x at
time step t, and Dist(x,y) is the number of cells
between cars x and y. According to the previous
rule, the velocity of each agent is increasing or
decreasing according to the state of the road. The
formulas are determined by: B A,

] 1 )) ( ) , ( ( [ = x v y x Dist Sign Sign A
t

t t t
w A x v w Sign Sign B y x = ] 1 )) ( ) ( ( [ ,
(5)
where function Sign(z) determines the sing of the
variable z such that,

=
0
, 1
, 1
) (z Sign
if
if
if
. 0
0
, 0
=
<
>
z
z
z

(6)

We will consider an alternate update of each
direction of motion, where there are synchronized
traffic lights at each road crossing site allowing
horizontal street motion at two-time steps and
vertical street motion at next two time steps. We
need four traffic lights: first time step is turn light to
allow the vertical motion, therefore cars a and b can
move along and also can turn right. Next time step,
the traffic light allows cars a and b to turn left. Third
time step is to the opposite case, i.e. horizontal
motion is allowed where cars c and d can go ahead
or turn right. Fourth traffic light is made for cars c
and d to turn left, and again these four time steps are
repeated.
Each of the N cars in the system starts at a randomly
selected site; its desired direction (left, straight on,
right) will be determined according to some rule
(discussed below). Each car is assigned a destination
site on the lattice. Once a car reached its destination
it will be assigned a new randomly chosen
destination. In the model of traffic system, the goal
of the car is the strategy: Drive as fast as you can
and slow down if you have to until arrive the
destination site! The goal of the system is always to
get a stable flow or high global average speed for all
cars. The driver is free to change its mind if an
intersection is momentarily locked. As a result, the
load on each road segment is well balanced and, as
long as there is a hole in the network, a motion will
occur. The traffic is not distributed uniformly and
some road segments are much more loaded than the
others. The load distribution changes with time, as a
result of the microscopic fluctuations.
In real traffic-flow systems, the traffic jam is
frequently induced by crossings when streets cross
with each other (crosscut road). The crosscut road
prevents cars from crossing their road. As soon as
crosscut road begins to be congested, a traffic jam
spreads from the crosscut road throughout space.
The crosscut road exists in the model where a finite
number of cars can wait approaching the crossing
from each of the four directions. Each car has a
desired direction, i.e. to the right, to the left or
straight on, which has to be chosen according to a
certain rule.
The behavior of cars to reach its destination in this
model is actually fairly easy to handcraft, while it is
hard to learn for many algorithms. Traffic
simulation here is made more realistic by given
individual drivers intentions, i.e. an idea of where
they want to go and choose an optimal trip. Let us
consider how we might develop the implementation
as a rough set technique. In our approach, the
decision tables define the behavior of the vehicle in
each state and are responsible for state changes.
Each car has an associated decision table that
defines the combinational actions in the behavior of
this car. This leads to dynamic knowledge bases, in
which the active sets of rules are determined by the
state of the agent.

Table 1: Condition and decision attributes for the
car in the system.
Attributes Description Value
s
(1) Relative_Spe
ed1
The velocity of
car at that
moment with
respect to the
{1,.
,k}

443
VOL. 2, NO. 9, September 2011 ISSN 2079-8407
Jour nal of Emer gi ng Tr ends i n Comput i ng and I nf or mat i on Sci ences

2009-2011 CIS Journal. All rights reserved.

http://www.cisjournal.org

444
Attributes Description Value
s
next car in front
of it
(2) Relative_Spe
ed2
The velocity of
car at that
moment with
respect to the
next car in left
road
{1,.
.k}
(3) Relative_spe
ed3
The velocity of
car at that
moment with
respect to the
next car in right
road
{1,
k}
(4) Move This car move at
that moment or
not
{Yes,
No}
(5) Jam The car exist in
jam road or not
{Yes,
No}
(6) Empty_next The next site has
car or not
{Yes,
No}
(7) Shortest_roa
d
Which road has
the shortest
distance to the
target
{left_r
oad,
right_
road,
front_
road}
(8) Longest_roa
d
Which road has
the longest
distance
{left_r
oad,
right_
road,
front_
road}
(9) Pass The car pass this
site during this
period
{Yes,
No}
(10) Time-late Travel-
time/expected-
time
{0, 1,
2, 3}
(11) D The decision
attribute
{go_st
raight,
turn_l
eft,
turn_r
ight}

The decision table of the car takes the form
, where U is the set of success time
steps, A is the condition attributes of the car (Table
1), and d is the decision attribute which takes values
of go straight, turn right or turn left. Basically, road
users want to get as fast as possible to their
destination, but not at all costs. They are also
interested in convenient transportation, which is
taken into account by several other criteria. There
exist dynamic and static road costs taken as
attributes in the decision table of each car. Dynamic
road cost is the travel time, which is determined on
every road online for each car. Static road cost is
road length where it is note that the shortest path is
not always the fastest. Generally, it is difficult to
determine an optimal road since there is a high
degree of subjectivity involved. For example roads
with low traffic density are usually perceived to be
faster than other ones, even if they are not, as the
mere feeling of getting somewhere is satisfying
driver. The conditional attribute jam in the Table
1 is determined by measure the velocity of the car, if
this velocity is a minimum value (v = 1) for more
than four time steps (because four time steps is the
maximum traffic light at crossing), then the car lies
in a jam. The condition attribute time-late can be
calculated using what called the expected time.
The expected time is defined as:
}) { , ( d A U
Expected time =
noise
velocity ected
ce dis ected
+
_ exp
tan _ exp

i.e.
Expected time noise
v
current et t
+

| arg |

so
Expected time = noise
v
j d i d
j i
+
+

| | | |
-----
-------------- (7)
and
t
v
v
t
i
i
=
=
1
-----------------(8)
Where d
i
and d
j
are the destination row and column
indices, i and j are the row and column indices of
the car site, v
i
is the velocity at time t =i, i >0, and
noise is a parameter. The possible values for
condition attribute time-late are:
0, which means early, If
, 8 . 0 _ exp / _ s time ected time travel
1, which means in the range, If
2 . 1 _ exp / _ 8 . 0 s < time ected time travel ,
2, which means some late, If
2 _ exp / _ 2 . 1 s < time ected time travel ,
and 3, which means it late, If
. 2 _ exp / _ > time ected time travel
Related to attribute time-late, If the car arrive the
target site and |expected_time travel_time| c s ,
where c is a parameter, then the car add all path-
traveling information to its decision table. The
attribute pass means that the car lost, or enters in a
loop, and then it cannot arrive its destination. So if
the car pass on the site more than twice in its way
to this destination, so the attribute takes value yes.
As we will see this attribute affects the decision, so
the car does not choose the same decision again.
VOL. 2, NO. 9, September 2011 ISSN 2079-8407
Jour nal of Emer gi ng Tr ends i n Comput i ng and I nf or mat i on Sci ences

2009-2011 CIS Journal. All rights reserved.

http://www.cisjournal.org
We will define the measure of distance between
current position and target one, which we will use to
calculate the values of attributes shortest-road and
longest-road. Let the position of the car is the site
and the target site is , then the
distance is defined as:
) , ( b a ) ' , ' ( b a

445
2 2
) ' ( ) ' ( b b a a dis + = .-------------(9)
When the driver needs to calculate the shortest or
longest road, he/she assigns to the next site
in each road, i.e. next site he/she will go in each
road (its road, right road or left road). The driver
goes straight when there does not exist any crossing
in front of the car. While at road crossing, the driver
use the decision table of the car to get set of rules
using rough set technique that describe the decision
which he/she should take at crossing of going
straight or turning. According to the rule satisfies
the conditions of the car at that moment, the driver
takes decision. If no rule satisfies the cars
conditions, then the decision of going on or turning
is taken randomly.
) , ( b a
The two-dimensional automata model consists
of sites on rectangular lattice with periodic
boundary conditions in both directions. We
simulated system of size L = 20, 30, 50, 100 and no
striking difference in the behavior of the system was
observed. Hence, we exclude finite size effects. The
fact that already comparatively small lattices exhibit
this independence from their size seems to indicate
the existence of only weak spatial correlations
between sites separated by a large distance.
L L
We observe some rules for number of cars that we
consider as good comparing with human expert
knowledge. We will illustrate new rules that differ
than in the previous phases as:
R
1
IF (move = no) AND (jam = yes) AND
(longest_road = right_road) THEN decision = left.
R
2
IF (relative_speed1 = 1) AND (shortest_road
= left_road) THEN decision = left.
R
3
IF (empty_next = yes) AND (longest_road =
left_road) THEN decision = right.
R
4
IF (short_road = left_road) AND (time_late =
1) THEN decision = left.
R
5
IF (move = no) AND (longest_road =
right_road) AND (pass = no) THEN decision = left.
R
6
IF (move = yes) AND (jam = no) AND
(longest_road = right_road) THEN decision =
go_straight.
R
7
IF (empty_next = yes) AND (jam = yes) AND
(time_late = 0) THEN decision = go_straight.
R
8
IF (move = yes) AND (pass = no) AND
(time_late = 0) THEN decision = go_straight.
R
9
IF (pass = no) AND (longest_road =
left_road) AND (time_late = 0) THEN decision =
go_straight.

Figure 3 shows the changing of average speed of
cars in the system with the density of cars. We
observe that the average speed of the cars reduced
with the increase of the density of cars in the
system. Figure 4 shows the changing of the flow in
the system with the density of cars. We observe also
that the flow of cars is relatively good in the high
density of cars.

0
0.5
1
1.5
2
2.5
3
3.5
4
0 0.2 0.4 0.6 0
A
v
e
r
a
g
e
S
p
e
e
d
D it fC
0
0.5
1
1.5
2
0 0.2 0.4 0.6 0.8 1
Density of Cars
F
l
o
w

o
f

C
a
r
s

Figure 4: The varying of flow with the
modification of density of cars.

Figure 3: The relation of total average speed and
density of cars.

We observe that the overall dynamics are quite
sensitive to the drivers behavior at road crossing for
choosing its destination. The emergent dynamical
property of driving control logic is defined in this
model clearly. It is possible to observe the emergent
behavior in this application of our new model such
that the emergent behavior in the traffic model lies
into two aspects one emerges from each other.
Firstly, the cars in the system emerge from the
cellular automata transition rule, where the
interactions between agents or cars produce the
movement of cars with different speed.




VOL. 2, NO. 9, September 2011 ISSN 2079-8407
Jour nal of Emer gi ng Tr ends i n Comput i ng and I nf or mat i on Sci ences

2009-2011 CIS Journal. All rights reserved.

http://www.cisjournal.org
6. CONCLUSION

The proposed model has provided a direct approach
to studying how dynamical systems perform
emergent computation; that is, how the interaction
of simple components with local information
storage and communication gives rise to coordinated
global information processing. Whether in real-life
situation, the topology of the interconnections that
gives meaning to the term immediate neighbor can
change frequently. Although every agent
participating in the system must be able to
communicate with its immediate neighbors, the
system itself should not depend on knowledge of the
overall system topology. The state transition within
each agent could be identical throughout the system
or unique to each agent. In practice, the state
transitions within the agents can most conveniently
be viewed as shared by all agents, but with local
adaptations as a function of either static or dynamic
local conditions.

REFERENCES

[1] Abul, O., Polat, F., and Alhajj, R., 2000,
Multiagent Reinforcement Learning Using Function
Approximation, IEEE Transaction on systems, man,
and cybernetics-part c: application and reviews, Vol.
30, No. 4, pp. 485-497.
[2] Darbha, S., Rajagopal, K., and Tyagi, V., A
review of mathematical models for the flow of
traffic and some recent results, Nonlinear analysis
69, 950-970, 2008.
[3] Gotts, N., 2000, Emergent Phenomena in Large
Sparse Random Arrays of Conways Game of
Life, International Journal of System Science, Vol.
31, No. 7, pp 873-894.
[4] Terrazas, G., Slepmann, P., Kendall, G., and
Krasnogor, N., An evolutionary methodology for
automated design of cellular automaton-based
complex systems, Journal of cellular automata, vol.
pp 77-102, 2007.
[5] Yasser Hassan and Eiichiro Tazaki, Emergence
Decision using Hybrid Rough Sets/Cellular
Automata, International Journal of System and
Cybernetics "Kybernetes", vol. 35, no. 6, pp. 797-
813, 2006.
[6] Hassan, Y., rough sets for adapting wavelet
neural networks as a new classifier system, Springer
applied intelligence journal, DOI: 10.1007/s10489-
010-0218-3, 2010
[7] Ngueyn, H.S., Skowron, A., and Szczuka, M.,
Situation Identification by Unmanned Aerial
Vehicle, In: Rough Sets and Current Trends in
Computing, (Ziarko, W., and Yao, Y. (eds.)),
Springer Verlag, pp. 49-56, 2002.
[8] Polkowski, L., Tsumoto, S., and Lin, Y., Rough
Set Methods and Applications, Physica Verlag,
2000.
[9] Tavakoli, Y., Javadi, H., and Adabi, S., A
cellular automata based algorithm for path planning
in multi-agent systems with a common goal,
International journal of computer science and
network security, Vol. 8, No. 7, 2008.
[10] Thangavel, K. and Pethalakshmi, A.,
Dimensionality reduction based on rough set theory:
a review, Applied soft computing 9m 1-12, 2009.

446