multi-objective evolutionary algorithms for energy...
TRANSCRIPT
MULTI-OBJECTIVE EVOLUTIONARY ALGORITHMS
FOR ENERGY-EFFICIENCY IN HETEROGENEOUS
WIRELESS SENSOR NETWORKS
José M. Lanza-Gutiérrez, Juan A. Gómez-Pulido, Miguel A. Vega-Rodríguez, Juan M. Sánchez
University of Extremadura (Spain). Polytechnic school
Index
1. Introduction
2. Heterogeneous wireless sensor network
3. Problem resolution
4. Experimental results
5. Comparisons with other authors
6. Conclusions and future work
2
08/02/2012
1. Introduction (I)
• The use of wireless sensor networks (WSNs) has
increased substantially in the last years:
applications in both civil and military areas.
• An important aspect in the use of WSNs is the
energy efficient. This kind of networks are
powered by batteries: network lifetime depends
on amount of information transmitted by sensors,
as well as its scope, among others.
• Nowadays, WSNs are more complex due to
inclusion of auxiliary elements (routers) in order
to minimize communication between sensors
,increasing both network speed and lifetime of
sensors Heterogeneous WSN
3
08/02/2012
1. Introduction (II)
• In this work, we have solved the
heterogeneous WSN design problem: how to
place routers and sensors optimizing
several objectives simultaneously.
• A good solution for heterogeneous WSN
involves an increase of energy efficiency
compared with its homogeneous equivalent.
• This is a NP-hard problem, so we need to use
certain techniques to facilitate its resolution,
like evolutionary algorithms. We have used
two well-known MOEAs: NSGA-II and
SPEA-2.
4
08/02/2012
1. Introduction (III)
Our work shows the following contributions:
1) The problem has been solved by means of evolutionary techniques.
2) We have optimized over three objectives that have not been considered jointly in any paper found: average number of hops, coverage and reliability.
3) The results obtained have been analyzed in depth using statistical procedures.
We study the deployment of a heterogeneous WSN as an alternative to traditional homogeneous WSN.
5
08/02/2012
2. Heterogeneous wireless
sensor network (I) A particular problem instance will be defined by
several elements: (N Routers, M sensors and a
sink node)
• A scenery (Dx * Dy)
• A sensor obtains information about its
environment with a sensitivity radius (Rs).
6
08/02/2012
2. Heterogeneous wireless
sensor network (II) • A router allows us to establish network
communications (links) and to collect
information about sensors in its
communication radius (Rc).
• Sink node collects information about all
sensors in the network, it is the center node.
• In this work, a sensor only can communicate
with routers, not with other sensors.
7
08/02/2012
2. Heterogeneous wireless
sensor network (III) The most important factors have been used to
deploy the network.
• Those that define the router network quality:
average number of hops (to minimize) and
reliability (to maximize).
• The global coverage provided by sensors (to
maximize).
These objectives are simultaneously optimized using
MOEAs.
8
08/02/2012
2. Heterogeneous wireless
sensor network (IV) Average number of hops (1): it is the minimum
number of hops (routers that are necessary to
cross) between each router and collector node,
divided by the total number of routers. A hop is
possible when the distance between two
elements is less than communication radius.
N number of routers, C collector node
9
08/02/2012
2. Heterogeneous wireless
sensor network (V) Sensor coverage (%) (2): it is the terrain
percentage covered by sensor nodes. We use a
boolean matrix of Dx*Dy points over scenery, so
for each sensor, the points within its radius will
be activated; finally, we have to count the
activated points.
R represents the boolean matrix and Rx,y the
position (x,y) of this matrix.
10
08/02/2012
2. Heterogeneous wireless
sensor network (VI) Reliability (%) (3): it allows us to define the
network robustness. It is the number of possible
paths between each router and collector node,
divided by the maximum number of paths in a
fully coupled topology.
TotalRoutes provides the number of paths
between two routers (Dijsktra). We notice that
when we use N+1 is because we have included
the collector node (N is the number of routers). 11
08/02/2012
3. Problem resolution (I) • The design of a heterogeneous WSN is a NP-
hard problem.
• It is necessary to use non-conventional
techniques to facilitate its resolution:
Heuristics, EAs,…
• We use MOEAs: the best results in literature.
• When we use this kind of techniques, there
are some important aspects: encoding of
individuals, crossover and mutation
strategies, generation of initial population.
12
08/02/2012
3. Problem resolution (II) EA is a search heuristic that mimics the process
of natural evolution. This heuristic is routinely
used to generate useful solutions to optimization
and search problems.
13
i Generation of initial
population (solutions)
F(x) Evaluation of
individuals
Se Selection of the best
individuals
Cr Crossover among
individuals.
Mu It alters values in a
individual
Re Worse individual die,
population size is constant
? Termination condition
08/02/2012
3. Problem resolution (III) Encoding of individuals: Two parts,
coordinates (x and y) of routers and sensors.
Each part is divided in regions to split the
available space in several portions, and to
ensure a good distribution of elements.
14
08/02/2012
3. Problem resolution (IV) Generation of initial population: We place
routers and sensor randomly.
Crossover: We only cross among elements from
a same region. The objective is that each region
can evolve separately.
Mutation: We perform random changes over
coordinates of elements. For each change, we
evaluate the individual.
If this change causes better fitness values will be
accepted; in the negative case change will be
discarded, back to previous coordinates.
15
08/02/2012
3. Problem resolution (VI) Well-known MOEAs: (Selector algorithms)
• NSGA-II
• SPEA-2
Well-described in literature [18][19].
16
08/02/2012
4. Experimental results (I) The instance data used in this work represent a
couple of scenarios of 100x100 and 150x150
meters, on which will be placed a set of routers
and sensors with values of Rc and Rs, 25 and 15
respectively (in meters). Providing of other
authors.
For these instances, we use the less number of
sensor nodes: scenario area divided by sensor
area.
17
08/02/2012
4. Experimental results (II)
We solve these 4 instances by both algorithms
NSGA-II and SPEA-2 and we obtain a solution
set (Pareto front) for each of them.
18
Instance A(m2) N M N/M
Inst1 100x100 4 16 4
Inst2 100x100 8 16 2
Inst3 150x150 4 32 8
Inst4 150x150 8 32 4
08/02/2012
4. Experimental results (III) To determine the goodness of solutions, we use
hypervolume metric (it is based on physical area
of this solution set).
If hypervolume is bigger, solution will be better.
Certain to ideal values (maximum coverage and
reliability (100%) and minimum number of hops
(0)).
19
08/02/2012
4. Experimental results (IV) By means of statistical techniques, we have
detemine that SPEA-2 provides better results
(hypervolumes) than NSGA-II for these
instances.
20
08/02/2012
5. Comparisons with other
authors (I) we can found results from resolution of
traditional WSN for energy efficiency, but we
cannot compare our fitness values with theirs
different conception.
Some authors have been demonstrated that
heterogeneous WSN provides better energy
efficiency than its homogenous alternative, but
their approaches are different from ours.
21
08/02/2012
5. Conclusions and future
work (II) In this work, we have tackled the deployment of
a heterogeneous WSN optimizing some
important factors: area covered by sensors,
average number of hops and network reliability.
We have used two well-known EAs, NSGA-II
and SPEA-2, proving as SPEA-2 provides the
best results.
Important: we have tackled how to obtain the
best heterogeneous WSN, but we have not
compare with its homogeneous conception.
22
08/02/2012
5. Conclusions and future
work (III) Future: more instances, new algorithms,
parallelism…
And a new approach, first, we study the
positioning of sensors maximizing coverage, and
then we deployed the network of routers
optimizing factors used in this work, including a
new metric for energy efficiency: allowing as
transform a real homogeneous WSN in a new
more energy efficiency heterogeneous WSN.
23
08/02/2012
MULTI-OBJECTIVE EVOLUTIONARY
ALGORITHMS FOR ENERGY-EFFICIENCY IN
HETEROGENEOUS WIRELESS SENSOR
NETWORKS
José M. Lanza-Gutiérrez, Juan A. Gómez-Pulido, Miguel A. Vega-Rodríguez, Juan M. Sánchez
University of Extremadura (Spain). Polytechnic school
Thanks for you attention