You are on page 1of 9

Swarm and Evolutionary Computation 1 (2011) 195203

Contents lists available at SciVerse ScienceDirect

Swarm and Evolutionary Computation


journal homepage: www.elsevier.com/locate/swevo

Regular paper

Energy-aware evolutionary routing protocol for dynamic clustering of wireless sensor networks
Enan A. Khalil, Baraa A. Attea
Department of Computer Science, Baghdad University, Iraq

article

info

abstract
The main challenges in designing and planning the operations of Wireless Sensor Networks (WSNs) are to optimize energy consumption and prolong network lifetime. Cluster-based routing techniques, such as the well-known low-energy adaptive clustering hierarchy (LEACH), are used to achieve scalable solutions and extend the network lifetime until the last node dies (LND). Also, evolutionary algorithms (EAs), have been successfully used in recent years as meta-heuristics to address energy-aware routing challenges by designing intelligent models that collaborate together to optimize an appropriate energyaware objective function. On the other hand, some protocols, such as stable election protocol (SEP), are concerned with another objective: extending the stability time until the first node dies (FND). Often, there is a tradeoff between extending the time until FND and the time until LND. To our knowledge, no attempt has been made to obtain a better compromise between the stability time and network lifetime. This paper reformulates the design of the most important characteristic of the EA (i.e., the objective function), so as to obtain a routing protocol that can provide more robust results than the existing heuristic and meta-heuristic protocols in terms of network stability period, lifetime, and energy consumption. An evolutionary-based routing protocol is proposed, which can guarantee better tradeoff between the lifespan and the stability period of the network with efficient energy utilization. To support this claim, extensive simulations on 90 homogeneous and heterogeneous WSN models are evaluated and compared against the LEACH, SEP, and one of the existing evolutionary-based routing protocols, hierarchical clustering-algorithm-based genetic algorithm (HCR). 2011 Elsevier B.V. All rights reserved.

Article history: Received 31 January 2011 Received in revised form 4 June 2011 Accepted 20 June 2011 Available online 1 July 2011 Keywords: Clustering Energy-aware Evolutionary algorithm Meta-heuristic Routing protocol Wireless sensor network

1. Introduction Wireless Sensor Networks (WSNs) have gained worldwide attention in recent years, particularly with the proliferation of Micro-Electro-Mechanical Systems (MEMS) technology, which has facilitated the development of smart sensors. WSNs are used in numerous applications, such as environmental monitoring, habitat monitoring, prediction and detection of natural calamities, medical monitoring, and structural health monitoring [14]. WSN is undergoing intensive research to overcome its complexity and constraint challenges in terms of storage resources, computational capabilities, communication bandwidth, and more importantly, power supply [5,6]. The main components of a sensor node and its associated energy consumption are depicted in Fig. 1. Typically, sensor nodes are grouped hierarchically in clusters (sections), and each cluster has a node that acts as the cluster head (CH). All the nodes forward their sensor data to the CH, which in turn aggregates data reports and routes them to a specialized node

Corresponding author. Tel.: +964 7801 614354. E-mail addresses: enanameen@yahoo.com (E.A. Khalil), baraaali@yahoo.com (B.A. Attea). 2210-6502/$ see front matter 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.swevo.2011.06.004

called the sink node or base station (BS). The low-energy adaptive clustering hierarchy (LEACH) protocol and the LEACH-Centralize (LEACH-C) [7,8] are well-known hierarchical routing protocols in WSN. They divide a WSN into dynamic clusters, each of which has a CH being elected according to a priori optimal probability. Although these LEACH-type schemes guarantee a welldistributed energy load, their analytical results are based on the assumption that the nodes of the sensor network are equipped with the same amount of energy, which is the case of homogeneous sensor networks. On the other hand, there are a lot of crucial WSN applications that would highly benefit from understanding the impact of equipping a percentage of the node population more energy than the rest of the networks nodes to ensure that the feedback result from the network be more reliable or stable. For example, the huge damage caused by the forest fire in Kentucky, USA, in 2007 (estimated to be multibillion US$) has motivated in developing a new WSN solution that can work in a harsh environment with heterogeneous data and devices. In addition, there are also applications where the spatial density of the sensors is a constraint. By assuming that with the current technology the cost of a sensor is tens of times greater than that of the embedded batteries, it will be valuable to examine whether the stability time (as measured by the period from the start of the network operation until the

196

E.A. Khalil, B.A. Attea / Swarm and Evolutionary Computation 1 (2011) 195203

2. Literature review 2.1. Heuristic protocols In LEACH [7,8], a node elected as CH with a probability p and each non-CH node determines its cluster by choosing the CH that can be reached with the least communication energy. The role of CH is rotated through all the sensors according to a random number T between 0 and 1. A node becomes a CH for the current rotation round if the number is less than the following threshold: T ( s) =

Fig. 1. Components of the sensor node and energy consumption for data aggregation and communication task.

p 1 p rmod 1 p 0 otherwise

if s G

first node dies FND) could be increased by simply distributing the extra energy to some existing nodes without introducing new nodes [9]. To this end, several heuristic protocols were proposed in the literature to cope with heterogeneity. The stable election protocol (SEP), then, weights the CH election probabilities according to the initial energy of a node relative to that of the other nodes in the network. SEP is proved to be more resilient than LEACH in judiciously consuming the extra energy of the advanced (more powerful) nodes [9]. Obviously, the clustering problem is considered as NP-hard, where the arena of computational intelligence (CI) is widely utilized for [1014]. Thus, different approaches of CI, including evolutionary algorithms (EAs), swarm intelligence (SI), and more recently, harmony search (HS), have been used as populationbased meta-heuristic optimization methods by different researchers for energy-aware cluster-based routing. In [1521], researchers drew attention to developing genetic cluster-based routing protocols in WSNs. While their protocols succeeded in prolonging the WSN lifetime (the time interval from the start of the network operation, until the last node dies LND) when compared with LEACH performance, they failed to ensure a longer reliability period until FND. Usually, there is a tradeoff between the time until FND and the time until LND. An optimal balance between these two contradictory goals is a challenge, and is itself an optimization process. Thus, fine tuning of the routing protocol parameters is essential to improve the overall network performance. By careful analysis of the genetic cluster-based routing protocols mentioned in [1521] and more specifically in the context of their major components, one can observe a common facet for the formalism of two energy-reduction criteria. In the objective function, the amount of energy can be mainly minimized based on two terms: network distance and cluster count. The formulization of the distance function has been addressed from an abstract perspective, where the smaller the distance between any two nodes, the lesser is the transmit power required to transmit the data. The cluster count has been formalized in the fitness function either directly or indirectly for the purpose of minimizing the total energy consumption. To our knowledge, no attempt has been made to get an acceptable compromise between the stability period and the network lifetime. This paper describes an evolutionary-based clustering protocol for the routing problem in WSNs, where all the transmissions are based on single-hop communication. The main goal of this work is the formulation of a new objective function that can have a significant impact on the overall performance of the WSN. It is tailored to meet the following conflicting goals: maximal stability period until FND, maximal network longevity until LND, while at the same time, minimizing energy consumption throughout the network lifetime. The rest of the paper is organized as follows. In Section 2, related work in both heuristic and metaheuristic evolutionary-based directions is reviewed. The details of the proposed protocol are discussed in Section 3. Simulation results are presented and discussed in Section 4, and concluding remarks are provided in Section 5.

(1)

where p is the desired percentage of the CH nodes in the sensor population, r is the current round number, and G is the set of nodes that have not been CHs in the last 1/p rounds. While LEACH-type schemes assume homogeneous WSNs, SEP [9] maintains hierarchical routing in heterogeneoussensor networks, where a percentage of the sensor population is equipped with extra energy than the rest of the normal sensors in the same network. If the fraction of advanced nodes is m and the additional energy factor between advanced and normal nodes is , then pnrm = p/(1 + m) padv = p (2) (3)

(1 + m)

(1 + ).

Hence, in SEP, the threshold in (1) is replaced by that for the normal sensors, T (snrm ), and that for advanced nodes T (sadv ) as follows [9]: T (snrm ) =

pnrm 1 pnrm rmod p 1 otherwise padv 1 padv rmod p 1

if snrm G

nrm

(4)

T (sadv ) =

if sadv G

adv

(5)

otherwise

where r is the current round, G is the set of normal nodes that have not become CHs within the last 1/pnrm rounds of the epoch, and T (snrm ) is the threshold applied to a population of n (1 m) normal nodes. This guarantees that each normal node will become a CH exactly once every 1/p (1 + m) rounds per epoch, and that the average number of cluster heads that are normal nodes per round per epoch is equal to n (1 m) pnrm . Similarly, G is the set of advanced nodes that have not become CHs within the last 1/padv rounds of the epoch, and T (sadv ) is the threshold applied to a population of n m advanced nodes. This guarantees that each advanced node will become a CH exactly once every (1/p) (1 + m)/(1 + ) rounds. 2.2. Meta-heuristic protocols As the clustering problem is proved to be an NP-hard problem, diverse fields of methods such as evolutionary computation are explored for proposing new algorithms. A detailed review of the evolutionary clustering algorithms is provided in [22], together with an investigation of application of Differential Evolution (DE) as robust, fast and fully automatic clustering technique that can circumvent the problems with the traditional clustering schemes. In WSNs, CI and its main branch EAs have been tailored or hybridized to suit several challenges. A detailed survey can be found in [23]. In the present study, a review of the field of the population-based meta-heuristic cluster-based routing in WSN is carried out with a focus on evolutionary algorithms. In [15],

E.A. Khalil, B.A. Attea / Swarm and Evolutionary Computation 1 (2011) 195203

197

the GA attempts to find appropriate CHs to minimize the total network distance. It uses binary individual, I, representation, in which each bit corresponds to one sensor node. 1 signifies that the corresponding sensor is a CH; otherwise, it is a regular node. The total transmission distance and the number of CHs Two are subjected to minimization in this GA using a combined fitness function GA .

wi , subsequently, these weights are updated according to the best


fit chromosomes. The fitness function of the GA in [21] includes, in addition to the above-mentioned five parameters, two more parameters concerning the remaining energy (RE) available in all the sensor nodes of the clusters and the number of frame transmissions (FT ), i.e., the frames received by the sink. The final fitness function can then be formulated as

GA (I ) = (TD D) + (1 ) (TN NCH )

(6)

where TD is the total distance of all the nodes to the BS, D is the sum of the distances from the regular nodes to CHs plus the sum of the distances from all the CHs to the BS, TN is the total number of nodes, NCH is the number of CHs, and w is a pre-defined weight. Except for D and NCH , all the other parameters are fixed values in a given WSN topology. The shorter the D, or the lower the number of NCH , the higher is the fitness value of an individual. This GA attempts to maximize the fitness value to determine a good solution. In [16], a genetic clustering algorithm (GCA) is proposed for the dynamic formation of clusters in WSN with the goal of increasing network longevity by minimizing energy dissipation. Each chromosome is represented as a fixed length list equal to the size of the network. The gene value can be 1 indicating that the corresponding node is dead, or a positive number referring to the node number of its CH. The fitness function, GCA , formulated as the minimization of (7), is composed of three parameters: number of CH nodes (NCH ), network Euclidean distance between all the nodes in each cluster to their CH (ND), and a system-dependent weight (w ) for adjusting both NCH and ND in calculating the fitness value for each chromosome, I.

GA (I ) =

(i , fi ),

fi {TD, D, E , SD, T , RE , FT }.

(9)

Other population-based meta-heuristic algorithms have also been developed for the cluster-based routing problem in WSN. Recently, the harmony search algorithm (HAS) [24] had been developed for improving the longevity and reducing the energy consumption in the clustered routing of WSNs [25]. The fitness of a harmony solution is defined as

HSA (Solution) = w f1 + (1 w) f2

(10)

where f1 is the maximum Euclidean distance of the nodes, defined as max{ nodei Ck d(nodei , CH k )/Ck }, and f2 is the ratio of the energy of all the alive nodes in the network with the total current energy of the CH in the current round, defined as N k i=1 E (nodei )/ j=1 E (CH k ). 3. The proposed energy-aware protocol In an attempt to harness the strength of the evolutionarybased routing protocol mentioned in Section 2 in terms of network stability period, longevity, and energy consumption, this section will revisit and elaborate the most important component that can constitute an effective evolutionary-based routing protocol, the fitness function, and reformulate it again in a new robust way. A centralized single-hop clustering protocol is presented where the BS runs an evolutionary-based protocol to optimize the CH election for cluster formation. To satisfy the EA conceptual development, this section will present the characteristics of the proposed EA-based protocols, viz. the energy-aware evolutionary routing protocol (EAERP), in both informal and formal ways. During the election phase, clusters are created using a centralized evolutionary algorithm. An initial population of individuals is generated and each individual is evaluated using a fitness function. Then, these individuals will go through evolutionary operators selection, recombination and mutation with pre-determined probabilities to improve the quality of the individuals. The evolutionary algorithm loop will continue until the termination criteria satisfied. In the context of the EAs, a complete clustered route solution is regarded as an individual, I. For a WSN with N sensor nodes, the individual consists of N genes, the alleles of which can be either 0 for non-CH nodes, 1 for CH nodes, or 1 for dead (i.e., inactive) nodes with no energy (E). Then, a population, I n , of n individual solutions can be formally specified as

GCA (I ) = NCH + (1 ) ND

(7)

In [1720], the authors extend the fitness function of [15] to include additional parameters, such as the standard deviation of the cluster distance (SD), the estimated transfer energy (E), and the number of transmissions (T ), as given in (8). These parameters ensure that the chromosomes (network configurations) that give a maximum number of transmissions and have reduced energy consumption are selected for future generations. As a result, their hierarchical clustering-algorithm-based genetic algorithm (HCR) with the extended fitness function, HCR , can provide more data transmissions when compared with the GA presented in [15].

HCR (I ) =

(i , fi ),

fi {TD, D, E , SD, T }

(8)

where the direct distance (TD) is computed as the sum of all the Euclidean distances from the sensor nodes to the BS. Cluster distance (D) is the sum of the distances from the nodes to the CHs and the distance from the heads to the BS. The variation in the cluster distances (SD) should be tuned according to the deployment information. If the deployment is uniform, the variation in cluster distances will be a strong indicator of optimum clusters; however, for random deployment, the variation in cluster distances will be a weak indicator. The transfer energy, E, represents the total energy consumed to transfer the aggregated message from the clusters to the sink. For a cluster with k member nodes, the cluster transfer energy (Ec ) sums up the energy required to transmit messages from k member nodes to the CH with the energy required by the CH to receive k messages from the member nodes and energy required to transmit aggregated data from the CH to the sink. Finally, for each data transfer stage, BS assigns the number of transmissions T . The value of T is adjusted according to the network conditions and current energy levels. Moreover, larger values of T indicate that the outcome of GA will be used for a longer period of time. While the initial fitness parameters are assigned arbitrary weights,

i {1, . . . , n} and j {1, . . . , N } 1 if E (nodej ) > 0 and nodej = CH i if E (nodej ) > 0 and nodej = non-CH Ij = 0 1 otherwise.

(11)

Each individual is randomly initialized with 1s and 0s according to the probability p, presented in (1), of the desired percentage of the CH nodes, as

Ij =

1 0

if E (nodej ) > 0 and randomj = p if E (nodej ) > 0 and randomj > p otherwise.

(12)

198

E.A. Khalil, B.A. Attea / Swarm and Evolutionary Computation 1 (2011) 195203

This genotype representation implicitly facilitates the formation of a dynamic number of CHs during the single and throughout the whole rounds of the routing protocol. Associated with each individual is a fitness (objective) value measured by a fitness function, , which numerically quantifies how good that individual is a solution to the routing optimization problem. It forms the bridge between the routing problem itself and the evolutionary algorithm. For EAERP, the proposed objective function is defined as the minimization of the total dissipated energy in the network, measured as the sum of the total energy dissipated from the nonCHs to send data signals to their CHs, and the total energy spent by CH nodes to aggregate the data signals and send the aggregated signals to the base station. Formally speaking, the fitness function used to evaluate individual, I k , k {1, . . . , n}, in EARP protocol becomes

{1, . . . , N 1}, and the participating parent individuals, I1 , I2 , are


then swapped at alleles between these two points as follows: R{pc } : I 2 I 2
I1 = (I1,1 , . . . , I1,r1 , I2,r1+1 , . . . , I2,r2 , I1,r2+1 , . . . , I1,N ) I2 = (I2,1 , . . . , I2,r1 , I1,r1+1 , . . . , I1,r2 , I2,r2+1 , . . . , I2,N ).

(17)

Each active allele in the new individuals is then mutated with the probability pm . Once an allele is chosen for mutation, its value is inverted from 0 to 1 and vice versa: M{pm } : I I

(i 1, . . . , n} and j {1, . . . , N }) : { I i if Iji = 1 or random > pm Iji = j 1 Iji otherwise.

(18)

EAERP (I ) =
k

nc i=1 sci

ETX s,CH + ERX + EDA


i

nc i =1

ETX CH ,BS
i

(13)

where nc is the total number of CHs, s ci is a non-CHs associated to the ith CH node, ETX node1,node2 is the energy dissipated for transmitting data from node1 to node2. Both the free space and multipath fading channel models mentioned in [79] will be used in this paper to compute the energy dissipated during the process of transmitting (ETx ) and receiving information (ERX ). For transmitting an l-bit message over a distance d, ETx is

In each round of the routing protocol, the cluster formation phase generates an initial population of solutions, the fitness of which is then evaluated and based on the fitness values, the parents are selected to generate a new population via recombination and mutation operators. This process is repeated until the termination condition of the evolutionary algorithm occurs. During the association phase, the phenotype of the best individual, best_I, among the k individuals in the population is considered as the clustering solution. best_I can be formally specified as I I n : EAERP (I ) EAERP (best_I ). (19)

ETX node1,node2

2 Eelec l + Efs l d(node1, node2) if d d0 = Eelec l + Emp l d(node1, node2)4 if d > d0 .

(14)

While ERx = Eelec l, where Eelec is the energy spent to operate the transceiver circuit set to 50 nJ/bit, and Efs and Emp are the energy expenditures for transmitting l-bit data to achieve an acceptable bit error rate and are dependent on the distance of transmission in the case of free space model and multipath fading model, respectively. The values of these two parameters are set to Efs = 10 pJ/bit/m2 and Emp = 0.0013 pJ/bit/m4 , respectively. If the transmission distance is less than the threshold d0 , the free space model is applied; otherwise, the multipath model is used. The threshold d0 is calculated by equating the two expressions in (14) at d = d0 : d0 =

Then in best_I, each non-CH determines the cluster to which it belongs by choosing the CH that requires the minimum energy consumption. Formally speaking, let nc best be the number of CHs elected by best_I, and let nodei , i {1, . . . N } E (nodei ) > 0 be an active non-CH, then

j, k {1, . . . , nc best } nodei Cj : d(nodei , CH j ) < d(nodei , CH k )


4. Simulation results

(20)

Efs /Emp .

(15)

Another parameter also taken into account is the data aggregation energy expenditure, which is set as Eda = 5 nJ/bit/message. The next component of the proposed EAs is the selection operator. It selects partners using binary tournament selection from the current population and transfers them to the mating pool for reproduction. To produce a mating pool of n parents, the binary tournament selects the best individual from two randomly selected individuals of the population set, and repeats this process n times. A formal definition of this selection operator, S : I 2 I , is as follows: let Ii,r1 , Ii,r2 , i {1, . . . , n} bet two individuals, and r1, r2 U {1, . . . , n} are two uniformly distributed random numbers from the set {1, . . . , n}, then Ii =

Ii,r1 Ii,r2

if EAERP (Ii,r1 ) = EAERP (Ii,r2 ) otherwise.

(16)

Recombination (R) and mutation (M) are the perturbation operators, which can alter the routing solutions found in the population. A proportion pc of pairs of parents in the selected population are chosen for recombination. For each pair of parents, two cut points, r1, r2, are randomly selected from the range

All the routing protocols (LEACH, SEP, HCR, and EAERP) are implemented in MATLAB. The simulations were performed on six groups of WSNs (WSNs#1, . . . , WSNs#6), each with 15 different playground topologies. According to the heterogeneity of the sensors, the simulations were performed on three types of WSNs. The first case assumes homogeneous sensor networks, (WSNs#1 and WSNs#4) while the second experiments (WSNs#2 and WSNs#5) assume heterogeneous sensor networks with advanced nodes of 10% and the third experiments (WSNs#3 and WSNs#6) assume 20% heterogeneity. Moreover, half of the simulations (WSNs#1, WSNs#2, and WSNs#3) assumes a center-located BS (i.e., the maximum distance of any node from the sink is about 70 m), while the second half of the experiments (WSNs#4, WSNs#5, and WSNs#6) assumes a corner-located BS (i.e., the maximum distance of any node from the sink is about 141.42 m). Overall, the simulation results presented herein have been averaged over 15 simulation runs for a total of 90 different WSNs. Each sensor network is composed of 100 sensor nodes deployed randomly with different distribution patterns in a playground of 100 m 100 m sensor field (much of the WSN literature assumes that the sensors will be randomly deployed). This signifies that the horizontal and vertical coordinates of each sensor are randomly selected between 0 and the maximum value of the dimension. For fairness in comparison, the characteristics of the networks and communication model used for the competent protocol simulations are made identical as illustrated in [79, 1521]. Moreover, as the effect of the proposed fitness function

E.A. Khalil, B.A. Attea / Swarm and Evolutionary Computation 1 (2011) 195203

199

Fig. 2. Average number of alive nodes versus rounds for WSNs#1.

Fig. 4. Average number of alive nodes versus rounds for WSNs#3.

Fig. 3. Average number of alive nodes versus rounds for WSNs#2.

Fig. 5. Average number of alive nodes versus rounds for WSNs#4.

(EAERP ) on the final performance of the algorithm is to be studied, all other evolutionary components, including selection, crossover, and mutation, of HCR, and EAERP are fixed to binary tournament selection, two-point crossover with pc = 0.6, and mutation with pm = 0.03, respectively. This permits investigation of the pure effect of the fitness formulation on the final performance of the implemented protocol. Furthermore, equal weights are used in the fitness functions of HCR. The population size, n, is taken as 20 and allowed to evolve for 20 generations. Finally, to benchmark our routing protocol against other well-cited protocols, the performance is compared in terms of the length of the stability period (the time interval from the start of the network operation until the first node dies, FND), network lifetime (the time interval from the start of the operation until the last node dies, LND), throughput as measured by the number of aggregated packets sent from CHs to BS, and the total energy left in the network. Additionally, the quantitative results that clarify the best performance values are given in bold. 4.1. Stability period and network lifetime The results depicted in Figs. 213 qualitatively capture the performance of the protocols for the six groups of WSNs. For the graphs depicted in Figs. 813, each protocols bar-pair measures the length of the period until FND (left bar) and the length of the period until LND (right bar). For the WSN groups with the centerlocated BS (i.e., WSNs#1, WSNs#2, and WSNs#3), the network stability period provided by EAERP is superior over all other protocols except SEP in one case, the heterogeneous WSNs#3.

Fig. 6. Average number of alive nodes versus rounds for WSNs#5.

EAERP extends the stable (i.e., reliable) region to about 10% and 20% when compared with that of LEACH and HCR, respectively. Against SEP, EAERP extends the stable region to about 10% for WSNs#1 and 3% for WSNs#2. On the other hand, SEP competes with EAERP with 3% gain for WSNs#3. However, one can observe that the additional gain provided by the stable heuristic of SEP protocol over EAERP can easily bewilder it on achieving the second objective, and can end up with unacceptable network lifetime results. With regard to the total network lifetime, EAERP is found to be very favorably against all other protocols. When compared with

200

E.A. Khalil, B.A. Attea / Swarm and Evolutionary Computation 1 (2011) 195203
2500

2000

1500

1000

500

0 LEACH SEP HCR EAERP

Fig. 11. Summary of FND and LND for WSNs#4.


3500 3000

Fig. 7. Average number of alive nodes versus rounds for WSNs#6.


2500 2500 2000 1500 1000 1500 500 1000 0 LEACH 500 SEP HCR EAERP

2000

Fig. 12. Summary of FND and LND for WSNs#5.


4000 3500

0 LEACH SEP HCR EAERP

Fig. 8. Summary of FND and LND for WSNs#1.


3000 4500 4000 3500 3000 2500 2000 500 1500 1000 500 0 LEACH SEP HCR EAERP 0 LEACH SEP HCR EAERP 2000 1500 1000 2500

Fig. 13. Summary of FND and LND for WSNs#6.

Fig. 9. Summary of FND and LND for WSNs#2.


4500 4000 3500 3000 2500 2000 1500 1000 500 0 LEACH SEP HCR EAERP

Fig. 10. Summary of FND and LND for WSNs#3.

LEACH, EAERP extends the network lifetime by 54%, 67%, and 70% for WSNs#1, WSNs#2, and WSNs#3, respectively. Furthermore,

when compared with SEP, EAERP gains 54%, 90%, and 83% additional rounds for the three above-mentioned WSN groups, respectively, and when compared with the second evolutionarybased protocol, i.e., HCR, EAERP gains about 15%, 20.4%, and 16%, respectively. With regard to the results of the second group of WSNs with a corner-located BS (i.e., WSNs#4, WSNs#5, and WSNs#6), one can observe the following: in the beginning, LEACH, SEP, and HCR maintain a better performance than EAERP, yielding longer stability periods (over EAERP, LEACH gains about 30%, SEP gains from 30 to 45%, and HCR shows slight gains of no more than 8%.) Subsequently, EAERP recovers its performance, overtakes LEACH, SEP, and HCR via slowing down the rate of node death throughout the network lifetime, while other protocols speed up the death rate substantially. On an average, EAERP gains 40%50%, 40%74%, and 14%18% when compared with LEACH, SEP, and HCR, respectively. This can be returned back owing to the following reason while examining the individual protocol runs. Although all protocols distribute their CHs uniformly over the sensing field, we found that the number of CHs elected by EAERP to be in the opposite region of the BS tends to be less than that elected by

E.A. Khalil, B.A. Attea / Swarm and Evolutionary Computation 1 (2011) 195203 Table 1 Average remaining energy over protocol rounds for WSNs#1 (with a maximum of 2271 rounds). %Rounds 10 20 30 40 50 60 70 80 90 100 LEACH 40.3711 30.7434 21.1195 11.4972 2.7247 0.0002 0.0 SEP 40.3764 30.7560 21.1338 11.5120 2.7340 0.0001 0.0 HCR 41.2637 32.5298 23.7940 15.0677 7.1228 2.1915 0.3431 0.0023 0.0 EAERP 42.2384 34.4738 26.7110 18.9478 11.2217 4.6397 1.1996 0.2014 0.0159 0.0

201

Table 3 Average remaining energy over protocol rounds for WSNs#3 (with a maximum of 4543 rounds). %Rounds 10 20 30 40 50 60 70 80 90 100 LEACH 40.7604 21.5177 8.1012 3.6840 0.0648 0.0 SEP 40.7738 21.5359 5.0052 0.8006 0.0029 0.0 HCR 42.5367 25.0850 11.3594 5.8076 2.6678 0.8447 0.1449 0.0033 0.0 EAERP 44.4882 28.9758 14.4425 7.9650 4.9536 2.3422 0.7713 0.1945 0.0147 0.0

Table 2 Average remaining energy over protocol rounds for WSNs#2 (with a maximum of 4351 rounds). %Rounds 10 20 30 40 50 60 70 80 90 100 LEACH 36.5559 18.1266 4.4469 2.0847 0.0870 0.0 SEP 36.6227 18.2436 2.3986 0.1115 0.0023 0.0 HCR 38.2891 21.5801 7.8332 3.0678 1.3183 0.4095 0.0899 0.0012 0.0 EAERP 40.1577 25.3134 11.0495 4.4203 2.7777 1.5286 0.5916 0.1556 0.0193 0.0

Table 4 Average remaining energy over protocol rounds for WSNs#4 (with a maximum of 2263 rounds). %Rounds 10 20 30 40 50 60 70 80 90 100 LEACH 39.7868 29.5728 19.3579 9.1509 1.4219 0.00063 0.000065 0.0 SEP 39.7861 29.5728 19.3640 9.1564 1.3978 0.0033 0.00006 0.0 HCR 40.0382 30.0718 20.1080 10.4513 3.3675 0.5426 0.0330 0.00098 0.00004 0.0 EAERP 40.4334 30.8663 21.3030 12.2647 5.4187 1.7576 0.3624 0.0253 0.0008 0.0

LEACH, SEP, and HCR. This has two contradictory impacts on the overall performance of the protocols manifested by the tradeoff between the dissipated energy of the individual nodes and the dissipated energy of the whole network. While EAERP, with its fitness function, attempts to find a better collection of clustered routes that minimizes the overall energy dissipated by the network to send, receive, and aggregate packets, a number of individual CHs may maintain heavy loads from their cluster members before forwarding the final aggregated packets to the BS. Over rounds and as these CHs have to send the aggregated packets to a distant BS, their individual energies will be expended earlier than in the other protocols, but in spaced interval fashion. However, the scenario of uniform distribution of CHs in LEACH, SEP, or HCR tends to slow down the death of individual nodes, causing the occurrence of FND to be later than that of EAERP, but when the nodes exhaust all their energies, they most probably exhaust in nearer intervals and rapid fashion. For example, closer examination to the results of homogeneous WSNs#4, we observe that EAERP takes about 407.2 rounds between FND and 40% dead nodes, while LEACH, SEP, and HCR take, on an average, 141.9, 151.1, and 328.4 rounds, respectively. 4.2. Energy consumption and throughput Again, more quantitative results can be extracted for evaluating the performance of the protocols. Tables 16 present the average remaining energy in the networks while rounds of the protocols proceed for the WSNs with center-located and corner-located BS, respectively. In these tables, the * marker indicates the case where the death of the whole networks is encountered before approaching the required number of rounds for the corresponding protocol. The results indicate that the most energy-saving protocol is EAERP, which performs better than all the other protocols. This additional utilization of the energy introduced into the system provided by EAERP, as compared with LEACH, SEP, and HCR can be returned back to the fact that EAERP has, in its fitness function (EAERP ), a direct manipulation of the total dissipated energy. On the

Table 5 Average remaining energy over protocol rounds for WSNs#5 (with a maximum of 3911 rounds). %Rounds 10 20 30 40 50 60 70 80 90 100 LEACH 37.3592 19.7019 5.4618 2.4482 0.4871 0.000469 0.0 SEP 37.3593 19.7172 3.5861 0.1060 0.00031 0.0 HCR 37.8130 20.6554 7.1496 2.9045 1.1915 0.2447 0.0192 0.000439 0.0 EAERP 38.5059 22.0881 9.0430 3.4443 1.4371 0.6230 0.1545 0.0251 0.0033 0.0

Table 6 Average remaining energy over protocol rounds for WSNs#6 (with a maximum of 4536 rounds). %Rounds 10 20 30 40 50 60 70 80 90 100 LEACH 39.5095 19.0350 7.2858 2.6276 0.0387 0.002 0.0 SEP 39.5182 19.0332 3.6027 0.3455 0.0126 0.0040 0.0 HCR 40.0008 20.2502 8.3916 3.9426 1.2499 0.1428 0.0213 0.0067 0.00008 0.0 EAERP 40.7849 21.9796 9.7085 4.3832 1.9448 0.6675 0.1664 0.0428 0.0191 0.0

other hand, HCR has additional parameters that indirectly handle the total network dissipated energy, while LEACH and SEP concern, in their main objectives, with distributing the role of CHs among all the sensor nodes according to probabilistic models that guarantee the election of each node as CH every epoch/sub-epoch. Moreover, the second group of results (Figs. 1419) presents the protocols throughput as measured by the total number of aggregated packets received at BS from the CHs until a specified percentage of the total network energy is dissipated. From these results, one can observe that EAERP significantly outperforms LEACH, SEP, and HCR. The results say that EAERP achieves better

202

E.A. Khalil, B.A. Attea / Swarm and Evolutionary Computation 1 (2011) 195203

Fig. 14. Throughput against dissipated energy for WSNs#1.

Fig. 17. Throughput against dissipated energy for WSNs#4.

Fig. 15. Throughput against dissipated energy for WSNs#2.

Fig. 18. Throughput against dissipated energy for WSNs#5.

Fig. 16. Throughput against dissipated energy for WSNs#3.

Fig. 19. Throughput against dissipated energy for WSNs#6.

throughput while maintaining better energy consumption. This can also be traced back to the positive impact of the fitness function (EAERP ). As EAERP keeps the sensor nodes alive for a longer number of rounds than in the other protocols, it can transmit more amounts of the aggregated packets to the BS. 4.3. Computational time When using Intel Core i5 CPU 2.27 GHz, EA-based protocols take additional time (albeit small) to run each round. This result comes naturally because these protocols handle more than one solution (as the case in LEACH and SEP) in each round. For 20 different individuals to be evolved in 20 generations, an EA-based protocol needs to process 400 alternative solutions at each round. To run each round, the average time needed for either LEACH or SEP ranges from about 0.03 to 0.13 s, whereas HCR takes 0.152 s

for a network containing 10100 nodes. However, EAERP takes 0.091.8 s to run each round for a network containing 10100 nodes. 5. Conclusion The formation of dynamic cluster-based routing in WSN has turned out to be an NP-hard problem, making it highly unlikely to develop a polynomial-time algorithm to compute an optimal clustered route. This paper has presented a new evolutionarybased dynamic cluster formation in WSN. The proposed EAERP with the formulation of the fitness function (EAERP ), when compared with other protocols (LEACH, SEP, and HCR), has been proven to be a meaningful way in deriving clustered routes with better tradeoff between network stability and network lifetime, while guaranteeing a well-distributed energy consumption.

E.A. Khalil, B.A. Attea / Swarm and Evolutionary Computation 1 (2011) 195203

203

Future research work needs to focus on exploring more complex routing models. For example, instead of using a singlehop routing method that is used in the proposed EAERP, it would be more efficient to formulate the problem as a multihop routing among the clusters for routing the data to the base station. Furthermore, additional heuristics may be studied and applied in the construction of the objective function and/or other EA components to provide more network stability or longevity periods. References
[1] MIT, Emerging technologies that will change the world, IEEE Engineering Management Review (Feb.) (2004) 2030. [2] C.S.R. Murthy, B.S. Manoj, Ad Hoc Wireless Networks: Architectures and Protocols, 1st ed., Prentice Hall, 2004, May 24. [3] K. Romer, F. Mattern, The design space of wireless sensor networks, IEEE Wireless Communications 11 (6) (2004) 5461. [4] I.F. Akyildiz, W. Su, Y. Sankarasubramaniam, E. Cayirci, A survey on sensor networks, IEEE Communications Magazine (Aug.) (2002) 102114. [5] J.N. Al-Karaki, A.E. Kamal, Routing techniques in wireless sensor networks: a survey, IEEE Wireless Communications 11 (6) (2004) 628. [6] H.M. Ammari, Challenges and Opportunities of Connected k-Covered Wireless Sensor Networks From Sensor Deployment to Data Gathering, in: Studies in Computational Intelligence, vol. 215, Springer, 2009. [7] W. Heinzelman, A. Chandrakasan, H. Balakrishnan, An application-specific protocol architecture for wireless microsensor networks, IEEE Transactions on Wireless Communications 1 (4) (2002) 660670. [8] W. Heinzelman, A. Chandrakasan, H. Balakrishnan, Energy-efficient communication protocol for wireless microsensor networks, in: Proceedings of the 33rd International Conference on System Science, HICSS00, Hawaii, USA, Jan. 2000, pp. 110. [9] G. Smaragdakis, I. Matta, A. Bestavros, SEP: a stable election protocol for clustered heterogeneous wireless sensor networks, in: Second International Workshop on Sensor and Actor Network Protocols and Applications, SANPA 2004, Boston MA, Aug. 2004.

[10] G.K. Venayagamoorthy, A successful interdisciplinary course on computational intelligence, IEEE Computational Intelligence Magazine 4 (1) (2009) 1423. [11] A. Engelbrecht, Computational Intelligence: An Introduction, 2nd ed., John Wiley & Sons, New York, USA, 2007. [12] A. Konar, Computational Intelligence: Principles, Techniques and applications, Springer, 2005. [13] X.S. Yang, Nature-Inspired Metaheuristic Algorithms, Luniver Press, 2008. [14] C. Blum, A. Roli, Metaheuristics in combinatorial optimization: overview and conceptual comparison, ACM Computing Surveys 35 (2003) 268308. [15] S. Jin, M. Zhou, A.S. Wu, Sensor network optimization using a genetic algorithm, in: Proceedings of the 7th World Multiconference on Systemics, Cybernetics and Informatics, 2003. [16] S. Mudundi, H.H. Ali, A new robust genetic algorithm for dynamic cluster formation in wireless sensor networks, in: Proceedings of Wireless and Optical Communications, Montreal, Quebec, Canada, May 2007. [17] S. Hussain, A.W. Matin, Hierarchical cluster-based routing in wireless sensor networks, in: IEEE/ACM International Conference on Information Processing in Sensor Networks, IPSN, 2006. [18] S. Hussain, A.W. Matin, O. Islam, Genetic algorithm for hierarchical wireless sensor networks, Journal of Networks (JNW) 2 (7) (2007) 8797. [19] S. Hussain, O. Islam, A.W. Matin, Genetic algorithm for energy efficient clusters in wireless sensor networks, in: Proceedings of the 4th International Conference on Information Technology: New Generations, ITNG, IEEE Computer Society, 2007, pp. 147154. April. [20] A.W. Matin, S. Hussain, Intelligent hierarchical cluster-based routing, in: Proceedings of the International Workshop on Mobility and Scalability in Wireless Sensor Networks, MSWSN, in: IEEE International Conference on Distributed Computing in Sensor Networks, DCOSS, June 2006, pp. 165172. [21] E.M. Shakshuki, H. Malik, Multi-agent-based clustering approach to wireless sensor networks, Int. J. Wireless and Mobile Computing 3 (3) (2009) 165176. [22] S. Das, A. Abraham, A. Konar, Metaheuristic Clustering, in: Studies in Computational Intelligence, vol. 178, Springer Verlag, 2009. [23] R.V. Kulkarni, A. Frster, G.K. Venayagamoorthy, Computational intelligence in wireless sensor networks: A survey, IEEE Communications Surveys & Tutorials 13 (1) (2011) 6896. [24] X.-S. Yang, Harmony search as a metaheuristic algorithm, in: Z.W. Geem (Ed.), Music-Inspired Harmony Search Algorithm Theory and Applications, 2009. [25] D.C. Hoang, P. Yadav, R. Kumar, S.K. Panda, A robust harmony search algorithm based clustering protocol for wireless sensor networks, in: IEEE International Conference on Communications Workshops, 2010.

You might also like