Академический Документы
Профессиональный Документы
Культура Документы
Abstract—The increasing in the demand for Wireless Sensor Network (WSN). The WSN is composed of sensor nodes, which
Networks (WSNs) has intensified studies which are dedicated to are built with a sensor board, a processor, a radio and a bat-
obtain more energy-efficient solutions, since the energy storage tery. These components allow the sensor to perform sensing,
limitation is critical in those systems. Additionally, there are other
aspects which usually must be ensured in order to get an accept- processing and communication tasks inside a coverage radius.
able performance of WSNs, such as area coverage and network When the network is in operation, the sensors collect data from
connectivity. This paper proposes a procedure for enhancing a phenomenon which is being analyzed, and disseminate them
the performance of WSNs: a multiobjective hybrid optimization toward the sink node, using multihop communication. In the lit-
algorithm is employed for solving the Dynamic Coverage and
Connectivity Problem (DCCP) in flat WSNs subjected to node erature, several works show the importance of WSNs and their
failures. This method combines a multiobjective global on-de- applications for monitoring of: wild life [1], structural health
mand algorithm (MGoDA), which improves the current DCCP [2], [3], mine tunnels [4], toxic organic compounds in the envi-
solution using a Genetic Algorithm, with a local on line algorithm ronment [5], etc.
(LoA), which is intended to restore the network coverage soon
Since the sensor nodes usually have to be cheap and com-
after any failure. The proposed approach is compared with an
Integer Linear Programming (ILP)-based approach and a similar pact they present limited energy storage and low processing and
mono-objective approach with regard to coverage, network life- communication capabilities [6]. The impossibility of recharging
time and required running time for achieving the optimal solution or replacing the node battery, specially in networks installed in
provided by each method. Results achieved for a test instance show regions of difficult access, imposes a serious constraint for the
that the hybrid approach presented can improve the performance
of the WSN obtaining good solutions with a considerably smaller designer: each node in the network has a limited lifetime, which
computational time than ILP. The multiobjective approach still cannot be extended.
provides a feasible method for extending WSNs lifetime with slight There are often many possible network configurations which
decreasing in the network mean coverage. are available for network setup. Each possible configuration is
Index Terms—Evolutionary algorithms, multiobjective op- defined by the state of the sensor nodes, which can be active or
timization, performance evaluation, wireless sensor networks inactive. An active node is able to transmit its own data, and data
(WSNs). received from other sensors, to the nodes which are inside its
communication radius. On the other hand, each inactive node is
I. INTRODUCTION maintained in an energy-saving state, in which it consumes very
low energy and can be woken up any time it is necessary.
control scheme, some nodes are scheduled to sleep (or change All decision vectors which are not dominated by any other deci-
to energy-saving state), while other ones continue to collect and sion vector of a given set are called nondominated regarding this
disseminate data to sink nodes [8]. Therefore, a fundamental set. Consequently, a Pareto-optimal solution is a vector which
problem in high density WSNs is to minimize the number of is not dominated by any other vector of the feasible set , and
active nodes to save energy, while keeping the area coverage the set of all Pareto-optimal solutions is the Pareto-optimal set
and the connectivity of nodes. . The image of the Pareto-optimal set in the objective func-
This paper presents a multiobjective hybrid approach for per- tion space is usually called the Pareto-front .
forming the density control in WSNs, which are subjected to
out-of-power failure in sensor nodes. The problem is modeled III. DENSITY CONTROL IN WSNS: A SHORT REVIEW
as a multiobjective optimization one, in which the design cri- The limited energy of WSN sensors is one of the most restric-
teria are set as minimizing the energy consumption and max- tive aspects for their employment. Several studies have been re-
imizing the coverage while keeping the network connectivity. cently proposed for performing the density control in WSNs, in
A decision-making procedure is applied to extract a single de- order to extend the network lifetime, at the same time that the
cision from the Pareto-set each time a reconfiguration proce- connectivity and coverage constraints are kept satisfied. Some
dure is needed. The approach is composed of a Multiobjective of these studies are briefly discussed in the remainder of this
Global on Demand Algorithm (MGoDA), which uses Nondom- section.
inated Sorting Genetic Algorithm 2 (NSGA-II) [9] for solving Meguerdichian and Potkonjak [10] model the coverage
the Dynamic Coverage and Connectivity Problem (DCCP), and problem using ILP. For solving the resulting problem, the
a Local on line Algorithm (LoA), which uses some deterministic authors propose the employment of greedy algorithms based
rules to fast restore the network coverage after some failure. on Set Coverage Theory.
The proposed multiobjective methodology is compared with Chakrabarty et al. [11] also deal with the coverage problem
an Integer Linear Programming (ILP) based approach and with using linear programming. The authors present a mathematical
a mono-objective hybrid algorithm with similar characteristics. model to minimize the cost of deploying heterogeneous nodes
The final solutions provided by these methods are evaluated in the monitoring area subjected to coverage constraints.
with regard to network lifetime, network coverage, energy con- Vieira et al. [12] and Meguerdichian et al. [13] propose algo-
sumption and computational required computational time for rithms based on graph theory and computational geometry for
obtaining such solutions. solving the coverage problem in WSNs.
This paper is structured as follows: Section II presents a Tilak et al. [14] propose a scheduling scheme which period-
short discussion about multiobjective optimization. Section III ically chooses a subset of nodes to perform the network tasks
presents a short study of the state-of-art in density control and sends the remaining nodes to energy-saving state. The ob-
methods for WSNs. The definition of the problem is presented jective considered in this work is to minimize the number of
in Section IV. The hybrid algorithm which is proposed here active nodes in each time period, complying with connectivity
for dealing with the DCCP is discussed in Section V. Finally, and coverage constraints.
Section VI presents the results obtained by the proposed Nakamura et al. [7] propose an ILP model to the multiperiod
approach and by two other ones in five instances with up to coverage and connectivity problem in flat WSNs and solve it
100 sensor nodes. with the commercial optimization package CPLEX [15]. The
model adds, to the coverage and connectivity constraints, a set
II. MULTIOBJECTIVE OPTIMIZATION of constraints regarding the node energy limits. Its goal is to
A general multiobjective optimization problem can be stated provide a solution which ensures the best coverage and connec-
as tivity for all periods of time considering the node battery ca-
pacity. Quintao et al. [16] compare the CPLEX solutions with
the ones achieved with an evolutionary algorithm (EA), consid-
(1) ering only the WSN coverage as the relevant design criterion.
Wu et al. [17] present a preliminary attempt of dealing with
in which is the decision vector, is the optimiza- multiple objectives in this kind of problem. The authors aggre-
tion parameter space, is the feasible set, which is com- gate three objectives (communication cost, path loss and de-
posed of the solutions which satisfy the problem constraints, tected signal energy level) into a single function, which is min-
is the vector of objective functions for imized by a genetic algorithm (GA). Although this approach
the problem and is the set of efficient points. presents advances with relation to the mono-objective ones, it
The set of efficient points (or Pareto-optimal set), which con- cannot be considered as a true multiobjective one, since it solves
stitutes the solution of the multiobjective optimization problem, a scalar problem, whose solution is a single optimal point. Even
consists of all decision vectors in which the corresponding ob- if those objectives were aggregated using a weighted sum, and
jective vector cannot be improved in any dimension without the weights were varied, creating multiple scalar problems with
degradation in another one. Given two decision vectors and different solutions, it would be still possible to show that this
, is said to dominate if, and only if approach would lose efficient solutions, which could be useful
to the designer [18].
Rajagopalan et al. [19] present another approach which in-
(2) tends to deal with WSN density control using multiobjective
MARTINS et al.: A HYBRID MULTIOBJECTIVE EVOLUTIONARY APPROACH FOR IMPROVING THE PERFORMANCE OF WSNS 547
activation energy of sensor node ; demand point; the constraint (7) imposes that a sensor node can
only cover a demand point if it is active; and the constraint (8)
maintenance power of the sensor node at the
ensures that the variable is binary
period of time ;
transmission power between the sensor nodes and (5)
, at the period of time ;
(6)
reception power of the sensor node at the period
of time ; (7)
energy consumed by node the node at the period
of time . (8)
binary variable which assumes 1 if the node The set of constraints (9)–(12) handles with the connectivity
covers the demand point at the period of time problem. The constraints (9) and (10) ensure that there is at least
, or 0 otherwise; one path between each active sensor node and the sink node. The
binary variable which assumes 1 if the demand constraints (11) and (12) allow only active nodes for building
point is not covered at the period of time , such a path
or 0 otherwise;
decision variable which assumes 1 if the arc
is in the path which connects (9)
the sensor node to the sink node at the period
of time , or 0 otherwise;
decision variable which assumes 1 if the sensor (10)
node is activated at the period of time , or 0
otherwise;
(11)
decision variable which assumes 1 if the sensor
node is active at the period of time , or 0
otherwise; (12)
penalty factor incurred by do not cover the demand Finally, the constraint (13) establishes the lower bound for the
point . energy consumed in each node at a period of time
The energy consumed by each node at the period of time
is given by (3). This energy is calculated based on the energy (13)
spent by the activation, maintenance, reception and transmission
operations The solution for this problem consists of finding the set
of sensor nodes which must be active at each
period of time . This solution can be generated periodically or
when it is necessary to redesign the WSN. It indicates which
nodes are responsible for covering each of the demand points,
while ensuring at least one path between the active sensors and
the sink node.
It should be noticed that this formulation has been stated, in
(3) references [7], [20], in the form of an ILP problem in order to
allow the usage of linear programming solvers like CPLEX or
The mono-objective ILP formulation that imposes full cov- GPLK (GNU Linear Programming Kit).
erage for the DCCP is presented next.
B. Multiobjective Statement
The objective function which has been modeled for mini-
mizing the energy consumed in the WSN at each period of time The problem statement described in Section IV-A has been
is shown in (4) reformulated as a multiobjective optimization problem, which
is handled by a multiobjective genetic algorithm.
The objective function shown in (4) has been rewritten, con-
(4) sidering that the optimization procedures that will be employed
no longer need a linear formulation. The new energy consump-
in which stands for the problem decision variable vector: tion function, which should be minimized, is shown in (14)
.
The constraints (5)–(8) address the coverage problem: the (14)
constraint (5) guarantees that at least sensor nodes cover each
MARTINS et al.: A HYBRID MULTIOBJECTIVE EVOLUTIONARY APPROACH FOR IMPROVING THE PERFORMANCE OF WSNS 549
in which: cost of the minimum path between the node and V. MULTIOBJECTIVE ONLINE HYBRID
the sink node at the period time . ALGORITHMS—MULTIONHA
The main idea of this function is to minimize the total energy
consumed in the network, in order to save energy for extending Mono-objective approaches for the DCCP problem often look
its lifetime. The function considers the energy required for acti- for networks which ensure connectivity, consuming low energy
vating the nodes and for keeping them working. Besides, a term and providing full coverage. Although it is an ideal condition,
which expresses the minimum cost between a sensor and the the full coverage is not necessarily a hard constraint. In several
sink node is included in the function.1 This parameter is adopted situations some noncoverage can be accepted in order to im-
for penalizing the sensor nodes in which the path for reaching prove the network lifetime.
the sink node is excessively expensive. The arcs which compose The multiobjective approach proposed in this paper allows
the graph in which the search is performed are those ones whose this type of tradeoff, since the full coverage constraint is relaxed
lengths are not higher than the maximum communication radius as a “noncoverage” objective function, (15), which should be
of a transmitter node. minimized jointly with the energy consumption function, (14).
Eq. (14) is an alternative way of representing the network en- The set of Pareto-optimal solutions which constitute the out-
ergy consumption. The main difference between this model and come of the multiobjective algorithm provides additional infor-
the one shown in (3) and (4) is that it is possible to include the mation to the designer: the tradeoff between energy consump-
cost of the minimum cost path in (14), since it is not necessary tion and network coverage.
to ensure that the function is linear. The inclusion of helps The Multiobjective On line Hybrid Algorithm (MultiOnHA),
in the topology design, since it indicates the routing pathways that is proposed in this paper for addressing the multiobjective
between the sensors and the sink node. Additionally, the mini- DCCP, combines a global strategy, which rebuilds the whole
mization of tends to reduce the energy spent for transmit- network, with a local strategy, which provides a solution of
ting data, since it is proportional to the distance between the lower computational cost for reestablishing the coverage and
nodes. connectivity in the neighborhood of a node that has failed.
In the multiobjective formulation, the constraint given in (5) The global strategy, which considers the full WSN in design,
is relaxed as an objective function, which aims to control the has the advantage of restructuring the network considering the
coverage of the network. This objective function, which is mod- energy situation of all sensor nodes. It makes possible to find the
eled as a “noncoverage” penalty function that should be mini- optimal solution for each situation. However, as a drawback, this
mized, is shown in (15) strategy is not scalable in time, and can require high computa-
(15) tional time for obtaining a solution.
On the other hand, the local strategy only considers the neigh-
in which: borhood of a failed node for finding a new solution. It reduces
the number of candidate solutions which are achievable by the
“noncoverage” penalty factor; algorithm, but it is able to offer a solution much faster than the
global algorithm.
number of demand points which are not covered by
It is important to emphasize that both optimization strategies
any sensor at the period of time ;
are executed at the sink node. This architecture has been chosen
penalty factor for the number of active sensors. because it is more efficient with regard to power consumption:
The purpose of this function is to allow the minimization the sensor nodes, which have limited processing and energy ca-
of the area which is not covered by the network. This is per- pabilities, become responsible for sensing, receiving and trans-
formed by finding the number of points which are not covered mitting data tasks only (no additional energy is required for
by any sensor inside the sensing area.2 Besides, the number of processing optimization algorithms). On the other hand, this ar-
sensor nodes which are active is also included in the function chitecture has the drawback of causing a delay time (latency),
in order to differentiate two networks with the same coverage: which is the time required for a sensor to detect the failure of a
if two networks cover the same number of demand points, the neighbor sensor, send this information through the network to
one which requires a lower number of sensors for ensuring such the sink node, and receive the solution for the problem (the in-
a coverage tends to be preferable. These two terms (number of dication of which sensor nodes should be activated for reestab-
covered points and number of active sensors) are joined using a lishing the coverage).3
weighted sum with weights and . Although this delay time is not null, it is not considered in
The relaxation of the coverage constraint as an objective func- the cases handled in this work, under the assumption that it is
tion is useful in the cases in which it is not necessary to maintain irrelevant when compared to the dynamics of the physical phe-
full coverage all the time: some of the coverage can be lost for nomena which are measured in most practical cases (tempera-
increasing the network lifetime. This suggests a tradeoff, cov- ture, pressure, humidity, etc.). However, the latency should not
erage versus lifetime, which cannot be addressed by the mono- be ignored in the cases in which the quantities can vary consid-
objective version of the problem. erably in short time periods.
1The cost of each arch is given by the transmission power between the nodes.
3The required computation time for performing such an optimization is not
2Such as described in the beginning of the section the sensing area is dis- considered as an extra delay time, since it would be required in any configuration
cretized in a finite set of equally spaced demanding points. of network information processing.
550 IEEE SENSORS JOURNAL, VOL. 11, NO. 3, MARCH 2011
(17)
(18)
MARTINS et al.: A HYBRID MULTIOBJECTIVE EVOLUTIONARY APPROACH FOR IMPROVING THE PERFORMANCE OF WSNS 551
Since the neighborhood of is usually small, it is possible to • Solution achieved by the proposed algorithm considering
evaluate all neighbor nodes and choose the one which minimizes (labeled as mult-0.70).
this function. • Solution achieved by the mono-objective hybrid algorithm
The main purpose of LoA is to reconnect all the sensors in proposed in [20] (labeled as mono).
to the sink node, by adding a single new active sensor. If it is not • Solution achieved by the commercial software Ilog
possible, the algorithm calculates the shortest path between the CPLEX [15] when solving the ILP statement shown in
nodes in and the sink node, with the weight of arcs defined Section IV (labeled as CPLEX)4. It should be noticed that
as shown in (17). In this case, all the inactive nodes which are the CPLEX was applied only for solving the ILP problem.
present in the shortest paths are activated in order to reestablish It is important to say that a NSGA-II version using Flip Muta-
the network coverage and connectivity. tion instead of DCM has been also tested. The results achieved
by this approach are not presented here since the algorithm has
C. The MultiOnHA been clearly outperformed by the one which employs DCM for
The structure of the multiobjective hybrid algorithm (Multi- performing mutation.
OnHA), which combines the MGoDA and LoA described early in The sensor nodes have been placed at random in a 50 m 50
this section, is shown in Algorithm 1. Initially, the MGoDA algo- m square, following an uniform probability distribution. Each
rithm is executed for finding an initial solution. Then, after each sensor node is based on the commercial device MICA2 [22],
node failure, the LoA is performed in order to try to restore the and has the following features:
WSN coverage and connectivity. In each simulation time unit, • Sensing radius: 15 m.
both the number of active nodes and the total power consump- • Communication radius: 25 m.
tion in the current period are compared with the number of active • Activation energy (AE): 5 mAh.
nodes and the power consumption in the previous period. If the • Maintenance energy (ME): 13 mAh.
differences obtained are higher than prespecified thresholds, then • Transmission energy (TE): 20 mAh.
MGoDA is executed for improving the current configuration. • Reception energy (RE): 2 mAh.
Some comments should be presented about the relationship
Algorithm 1 MultiOnHA between the geometry of the area to be covered, the location of
the nodes, and the role that will be played by each node. The
inputs: set of sensor nodes , sink node , set of arcs
problem of node density control appears in situations where a
which connect the sensor nodes, set of arcs which connect
the sensor nodes to the sink node, set of demand points , large number of nodes are placed within the area to be covered,
sensing radius and minimum admissible coverage with some redundancy of nodes covering the demand points.
Usually, the sensors of the network which are near to the sink
1: procedure are used for routing and sensing, and the nodes which are far
2: ; initial WSN away from the sink are used only for sensing (they are likely to
be leaves on the routing tree). This means that the energy man-
3: while do
agement for improving the network timelife is likely to have, as
4: ; returns the index of a fail node bottleneck nodes, the ones nearby the sink node and, in some
when a failure occurs or otherwise cases, also the nodes within some critical paths which are the
5: if then only alternatives for carrying the information coming from large
portions of the network. This is the pattern of the end of lifetime
6: ;
phenomena which appears in the examples presented in this sec-
7: ; tion: the network ceases of satisfying the coverage constraints
8: if then when the nodes that are near to the sink run out of energy.
The genetic algorithm, which is executed inside MGoDA, has
9: been tuned with the following parameters:
10: end if • Population size: 300 individuals.
• Number of generations: 150 generations.
11: end if
• Crossover probability: 0.90 per pair.
12: end while • Mutation probability: 0.20 per individual.
13: end procedure A detailed description about the mono-objective hybrid algo-
rithm which is used for comparison can be seen in [20]. Since it
VI. RESULTS considers the coverage as a hard constraint it looks for solutions
which keep the coverage as close from 100% as possible.
Tests have been performed for several network instances, with
The commercial optimization package CPLEX has been used
different sizes. The data of the instances considered here can be
for solving the mono-objective ILP formulation of the problem
found in [21]. In this paper, an instance with 100 sensor nodes
shown in Section IV. Since CPLEX uses a deterministic enu-
is presented for illustrating the performance of the proposed ap-
merative method, it reaches the global optimum for the mono-
proach. Four solutions have been compared:
objective problem. However, the required running time by the
• Solution achieved by the proposed algorithm considering
(labeled as mult-0.95). 4For mono and CPLEX it is assumed mn = 1:00.
552 IEEE SENSORS JOURNAL, VOL. 11, NO. 3, MARCH 2011
TABLE I
100 SENSOR NODES—RESULTS SUMMARY
REFERENCES
[1] A. Mainwaring, D. Culler, J. Polastre, R. Szewczyk, and J. Anderson,
“Wireless sensor networks for habitat monitoring,” in Proc. ACM Int.
Work. Wireless Sensor Networks and Applications (WSNA’02), New
York, 2002, pp. 88–97.
[2] D. Mascarenas, E. Flynn, C. Farrar, G. Park, and M. Todd, “Powering
and interrogation of structural health monitoring sensor networks,”
IEEE Sensors J., vol. 9, pp. 1719–1726, 2009.
[3] Q. Ling, Z. Tian, Y. Yin, and Y. Li, “Localized structural health moni-
toring using energy-efficient wireless sensor networks,” IEEE Sensors
J., vol. 9, no. 11, pp. 1596–1604, Nov. 2009.
[4] H. Jiang, L. Chen, J. Wu, S. Chen, and H. Leung, “A reliable and high-
bandwidth multihop wireless sensor network for mine tunnel moni-
toring,” IEEE Sensors J., vol. 9, no. 11, pp. 1511–1517, Nov. 2009.
[5] F. Tsow, E. Forzani, A. Rai, R. Wang, R. Tsui, S. Mastroianni, C.
Knobbe, A. J. Gandolfi, and N. J. Tao, “A wearable and wireless sensor
system for real-time monitoring of toxic environmental volatile organic
compounds,” IEEE Sensors J., vol. 9, no. 12, pp. 1734–1740, Dec.
2009.
[6] S. Park, A. Savvides, and M. B. Srivastava, “Simulating networks
Fig. 4. Computation time for 36, 49, 64, 81, and 100 sensor node instances. of wireless sensors,” in Proc. Conf. on Winter Simulation (WSC’01),
Washington, DC, 2001, pp. 1330–1338.
[7] F. G. Nakamura, F. P. Quintao, G. C. Menezes, and G. R. Mateus, “An
optimal node scheduling for flat wireless sensor networks,” in Proc.
Fig. 4 shows the computation time of the three methods in the IEEE Int. Conf. Networking (ICN’05), 2005, vol. 3420, pp. 475–483.
five instance sizes considered (observe that the axis represents [8] J. Podpora, L. Reznik, and G. V. Pless, “Intelligent real-time adaptation
the logarithm of the running time). It is easy to note that the for power efficiency in sensor networks,” IEEE Sensors J., vol. 8, no.
12, pp. 2066–2073, Dec. 2008.
CPLEX computational time increases nearly exponentially with [9] K. Deb, A. Pratap, S. Agarwal, and T. Meyarivan, “A fast and elitist
the instance size. This behavior can render this approach useless multiobjective genetic algorithm: NSGA II,” IEEE Trans. Evol.
for large instance sizes. On the other hand, the computation time Comput., vol. 6, pp. 182–197, 2002.
[10] S. Meguerdichian and M. Potkonjak, Low power 0/1 coverage and
increases much slower for the mono-objective hybrid algorithm scheduling techniques in sensor networks Univ. California, Los An-
and the MultiOnHA. Therefore, it is expected that those methods geles, CA, Tech. Rep. 030001, 2003.
will still work even for very large instances. [11] K. Chakrabarty, S. S. Iyengar, H. Qi, and E. Cho, “Coding theory
framework for target location in distributed sensor networks,” in Proc.
Int. Symp. Informa. Technol.: Coding and Comput. (ITCC’01), 2001,
VII. CONCLUSION pp. 130–134.
[12] M. Vieira, L. Vieira, L. Ruiz, A. Loureiro, A. Fernandes, and J.
This paper proposes a multiobjective hybrid algorithm for Nogueira, “Scheduling nodes in wireless sensor networks: A Voronoi
approach,” in Proc. Ann. IEEE Int. Conf. on Local Computer Networks
performing density control in flat WSNs. The algorithm is in- (LCN’03), 2003, pp. 423–429.
tended to find solutions which ensure network connectivity and [13] S. Meguerdichian, F. Koushanfar, M. Potkonjak, and M. B. Srivastava,
keep the coverage above a certain level, while extending its “Coverage problems in wireless adhoc sensor networks,” in Proc. IEEE
Conf. Comput. Commun. (INFOCOM’01), 2001, pp. 1380–1387.
lifetime. The method combines a global on demand algorithm, [14] S. Tilak, N. B. Abu-Ghazaleh, and W. Heinzelman, “Infrastructure
which redesigns the whole network when it is required, with tradeoffs for sensor networks,” in Proc. ACM Int. Work. Wireless Sensor
a local on line algorithm, which tries to reestablish the net- Networks and Appl. (WSNA’02), 2002, pp. 49–58.
[15] Ilog CPLEX Source, 2006. [Online]. Available: http://www.ilog.com/
work coverage and connectivity always when a sensor fails. The products/cplex/
choice for executing one or other of the algorithms is based [16] F. Quintao, F. Nakamura, and G. R. Mateus, “Evolutionary algorithm
on the number of new active sensors and on the energy con- for the dynamic coverage problem applied to wireless sensor networks
design,” in Proc. IEEE Congr. Evol. Comput. (CEC’05), 2005, vol. 2,
sumption in the network (the local algorithm can be useful for pp. 1589–1596.
small changes, but it usually provides rough solutions when the [17] Q. Wu, N. Rao, J. Barhen, S. Iyengar, V. Vaishnavi, H. Qi, and K.
number of new active nodes is high). Chakrabarty, “On computing mobile agent routs for data fusion in dis-
tributed sensor network,” IEEE Trans. Knowl. Data Eng., vol. 16, pp.
The proposed MultiOnHA seems to be efficient for solving 740–753, 2004.
the DCCP problem in flat WSNs. The employment of a multi- [18] V. Chankong and Y. Y. Haimes, Multiobjective Decision Making:
objective approach has provided to the designer the additional Theory and Methodology. Amsterdam, The Netherlands: North-Hol-
land, 1983.
choice of losing some coverage in order to improve the network [19] R. Rajagopalan, C. Mohan, P. Varshney, and K. Mehrotra, “Multi-ob-
lifetime. This additional degree-of-freedom constitutes a feature jective mobile agent routing in wireless sensor networks,” IEEE Trans.
of MultiOnHA algorithm that is not found in any former de- Knowl. Data Eng., pp. 1730–1737, 2005.
[20] F. V. C. Martins, F. G. Nakamura, F. P. Quintao, and G. R. Mateus,
sign algorithm. Results achieved for a test instance have shown “Model and algorithms for the density, coverage and connectivity con-
that the MultiOnHA can obtain solutions which are comparable trol problem in flat WSNs,” in Proc. Int. Network Optimization Conf.
to the ones achieved by an exact method, spending a consider- (INOC’07), 2007, pp. 1145–1152.
[21] F. V. C. Martins, E. G. Carrano, E. F. Wanner, R. H. C. Takahashi,
ably smaller computational time. It was also noted that it is pos- and G. R. Mateus, A hybrid multiobjective evolutionary approach for
sible to extend the operation time of the network considerably, improving the performance of wireless sensor networks: simulation
by reducing the minimum admissible coverage. The ILP and data UFMG, 2010. [Online]. Available: http://www.mat.ufmg.br/ taka/
techrep/wirsens01.pdf
mono-objective hybrid approaches cannot provide such flexi- [22] Mica2— Wireless Measurement System XBOW, 2006. [Online].
bility in the network design. Available: http://www.xbow.com/
554 IEEE SENSORS JOURNAL, VOL. 11, NO. 3, MARCH 2011
Flávio V. C. Martins was born in Conselheiro Pena, Brazil, in 1982. He re- Ricardo H. C. Takahashi (M’95) was born in Ipatinga, Brazil, in 1965. He
ceived the B.S. degree in computer science in 2007 and the M.S. degree in received the B.S. and M.S. degrees in electrical engineering from the Univer-
electrical engineering in 2009 from the Universidade Federal de Minas Gerais sidade Federal de Minas Gerais (UFMG), Belo Horizonte, Brazil, in 1989 and
(UFMG), Belo Horizonte, Brazil. Since 2009, he has been working towards the 1991, respectively, and the Ph.D. degree in electrical engineering from the Uni-
Ph.D. degree in electrical engineering at the UFMG. versidade de Campinas, São Paulo, Brazil, in 1998.
From 2008 to 2009, he was with Centro Federal de Educação Tecnológica From 1992 to 2002, he was with the Department of Electrical Engineering,
de Minas Gerais (CEFET-MG), Belo Horizonte, Brazil, as a substitute Pro- UFMG, and he has been with the Department of Mathematics, UFMG, since
fessor. His current research interests include wireless sensor network, network 2002. He is the author of more than 50 journal papers and 100 conference pa-
design, evolutionary algorithms, combinatorial optimization, dynamic systems, pers, and has advised 12 Ph.D. theses. His main research interests include the
and multiobjective optimization. fields of optimization theory (including evolutionary computation and multicri-
terion optimization), computational intelligence, and optimization-based con-
trol theory.
Elizabeth F. Wanner received the B.S. degree in mathematics in 1994 and the
M.Sc. degree in pure mathematics in 2002, and the Ph.D. degree in electrical
engineering from the Universidade Federal de Minas Gerais (UFMG), Brazil,
in 2006.
She is currently an Assistant Professor with the Department of Computer En-
gineering, Centro Federal de Educação Tecnológica de Minas Gerais, Belo Hor-
izonte, Brazil. Her current research interests include evolutionary computation,
global optimization, constraint handling techniques for evolutionary algorithms,
and multiobjective optimization.