CCECE 2014 1569888203 Coverage Maxmzaton n Moble Wreless Sensor Networs Utlzng Immune Node Deployment Algorthm Mohammed Abo-Zahhad, Sabah M. Ahmed and Nabl Sabor Electrcal and Electroncs Engneerng Department Faculty of Engneerng, Assut Unversty Assut, Egypt (zahhad@yahoo.com, sabahma@yahoo.com, nabl_sabor@yahoo.com) Abstract A Wreless Sensor Networ (WSN) conssts of spatally dstrbuted autonomous sensors wth sensng, computaton and wreless communcaton capabltes. Each sensor generally has the tas to montor, measure ambent condtons, and dssemnate the collected data towards a base staton. One of the ey ponts n the desgn stage of a WSN that s related to the sensng attrbute s the coverage of the sensng feld. The coverage ssue n WSNs depends on many factors, such as the networ topology, sensor sensng model, and the most mportant one s the deployment strategy. The sensor nodes can be deployed ether determnstcally or randomly. Random deployment of the sensor nodes can cause coverage holes formulaton; therefore, n most cases, random deployment s not guaranteed to be effcent for achevng the requred coverage. In ths case, the moblty feature of the nodes can be utlzed n order to maxmze the coverage. Ths s Non-determnstc Polynomal-tme hard (NP-hard) problem. So n ths paper, the Immune Algorthm (IA) s used to relocate the moble sensor nodes after the ntal confguraton to maxmze the coverage area wth the movng dsspated energy mnmzed. The performance of the proposed algorthm s compared wth the prevous algorthms usng Matlab smulaton. Smulaton results show that the proposed algorthm mproves the networ coverage and the redundant covered area wth mnmum movng consumpton energy. I. INTRODUCTION In recent years, Wreless Sensor Networs (WSNs) have evolved very rapdly and have found many new applcatons. For example, WSNs can be used to montor the envronment, trac targets on a battlefeld or measure traffc on roads [1]. One of the ey ponts n the desgn of WSN that s related to the sensng attrbute s the coverage of the sensng feld. In the lterature [2], the coverage problem n WSNs has been studed ether as target coverage or area coverage. The area coverage algorthms are used to maxmze the covered area of the sensng feld [4]-[5] and [7]-[8]. On the other hand, the target coverage algorthms are adopted to maxmze the number of targets that could be covered based on assumpton that the sensng feld s dvded nto targets [3], [6]. The most mportant factor that has nfluence on the coverage problem s the deployment of sensor nodes n the sensor feld. The best deployment maxmzes the coverage area and also prolongs the operatonal lfe of the ndvdual nodes. The sensor nodes can be deployed ether determnstc or random. Random Shgenobu Sasa Department of Electrcal and Electronc Engneerng Ngata Unversty 8050 Iarash 2-no-cho, Nsh-u, Ngata 950-2181, Japan (ojro@eng.ngata-u.ac.jp). deployment s usually preferred n large scale WSNs not only because t s easy and less expensve but also t mght be the only choce n hostle envronments. However, random deployment of the sensor nodes can cause holes formulaton; therefore, n most cases, random deployment s not guaranteed to be effcent for achevng the maxmum coverage [3]-[5]. The paper s organzed as follows. Secton II s a lterature survey about varous deployment algorthms. The sensng model that s used n the proposed algorthm s descrbed n secton III. Secton IV explans the proposed mmune node deployment algorthm and how the IA s used to maxmze the covered area and mnmze the energy consumpton durng the movement process. In Secton V, the smulaton results and dscusson are gven. Fnally, secton VI offers some conclusons. II. RELATED WORK In order to overcome the problem of holes formulaton after ntal deployment of the sensor nodes n the sensng feld, an effcent algorthm that would maxmze the covered area or targets should be employed. In random deployment, there are two approaches to reduced or elmnated holes formulaton problem after ntal deployment. In the frst approach, f the sensor nodes are hybrd n whch some of the nodes are statonary and the other are moble, an evolutonary algorthm, such as genetc algorthm, should be employed n order to fnd the number and locatons of the moble nodes that should be added after the ntal deployment of the statonary nodes as descrbed n [3]. In ths method, the moble nodes move large dstance to acheve maxmum coverage and ths ncreases the movng consumpton energy. In the second approach, f all sensor nodes are moble, then an evolutonary algorthm should be desgned to fnd the locatons of all moble nodes based on the coverage maxmzaton [4]-[5], [7]-[8]. The author of [4] uses Partcle Swarm Optmzaton (PSO) approach to maxmze the coverage based on a probablstc sensor model n moble sensor networs. But, the movng energy consumpton s not consdered. A Two Phase PSO Algorthm s presented n [7] to enhance the networ coverage and reduces the movng energy consumpton. Both objectves are acheved n separate phases wth coverage maxmzaton n the frst phase whle energy conservaton n the second phase. 978-1-4799-3010-9/14/$31.00 2014 IEEE CCECE 2014 Toronto, Canada 1
Coverage problem s caused by lmted sensng range of sensor nodes. The soluton of ths problem depends on how the sensors are postoned wth respect to each other. Ths s Nondetermnstc Polynomal-tme hard (NP-hard) problem [9]-[10]. Ths paper presents one of the mult-objectve evolutonary algorthms that s devoted to solve the above mentoned problem; namely the Immune Algorthm (IA). IA s a randomzed algorthm nspred by mmune functons and prncples observed n nature [11]-[13]. It s used here to relocate the sensor nodes after the ntal confguraton based on the maxmzaton of the coverage area and mnmzaton of the dsspated energy durng the movement process. III. THE SENSING MODEL In the proposed algorthm, the followng assumptons about sensor nodes are fxed: Each sensor s coverage s a crcle. All sensors here have the same coverage wth radus Rs. All sensors have GPS or other locaton determnaton devces and can move to other postons wthn ther moblty range. Sensors cannot sense through or move across boundares and obstacles that are consdered walls. The qualty of sensng s constant wthn Rs and s zero outsde the sensng range,.e. t follows a bnary model. A. Coverage Problem The frst objectve consdered n the proposed algorthm s the coverage problem. The sensor model here s a bnary model, whch s supposed to be covered as much as possble. Ths means that the area wthn the sensng range can be counted as covered wth a probablty of 1 and the area out of the sensng range wll be set as 0 snce t cannot be covered. The sensng feld s consdered to be grds and each grd sze s equal to 1 as shown n Fg. 1. The coverage of the whole area s proportonal to the grd ponts that can be covered. Consderng the grd pont G(x, y), the possblty that t can be sensed by a sensor node s (x, y ) s descrbed by [5]-[7]: (,, ) = 1, ( ) + ( ) (1) 0, h Assumng that a WSN conssts of N moble sensors (.e. = {,,, }), the probablty that a pont G(x, y) s covered can be wrtten as: (,, ) = 1 (1 (,, )) (2) It should be ponted out that the area covered by each sensor s =, so, the total coverage area s gven by: ( ) = (,, ) (3) and the uncovered area s ( ) = ( ). As a result, the percentage coverage area s gven by: ( ) = ( ) (,, ) (4) where, A tot s the total area of the sensng feld. The man am of the proposed algorthm s the maxmzaton of the coverage area by mnmzng the uncovered area rato ( 1 ( ) ) as follows: ( = 1 ( )) (5) The probablty that the sensng range of the sensor node s and the sensor node s j to overlap on the grd pont G(x, y) as shown n Fg. 1 s gven by: 1, (, ) (, ), (,, ) =, [1, ], (6) 0, h The percentage redundant covered area of the Moble Wreless Sensor Networ (MWSN) n the sensor feld s gven by: ( ) (,, ) (7) Fg.1.Sensor coverage n sensng feld. B. Movng Energy Consumpton of Moble Nodes The second objectve consdered n the proposed algorthm s the mnmzaton of the movng energy consumpton. It s defned as the energy used for redeployng the sensor nodes. Ths has been carred out by mnmzng the movng energy consumpton of the sensor nodes through mnmzaton of the root mean of the sum squared moved dstances of all sensor nodes ( ) as follows: = (8) where, s the sensng range (.e. radus of coverage crcle) and s the root mean of the sum squared moved dstances of all sensor nodes. For N sensor nodes, s gven by: = ( ) (9) where, = ( ) + ( ) and (, ) and (, ) are the ntal and fnal postons of the th sensor node respectvely. 2
IV. THE PROPOSED NODES DEPLOYMENT ALGORITHM Intally, the sensor nodes are deployed randomly. Then, the base staton (BS) sends a short message to request the IDs, and ntal postons of all sensor nodes n the sensor feld. Based on the feedbac nformaton from sensor nodes, the BS uses the proposed mmune deployment algorthm to fnd new locatons of the MWSN based on the maxmzaton of the coverage area whle at the same tme the movng cost of the moble nodes s mnmzed. Fg.2 llustrates the flow chart of the proposed algorthm. The man steps of the proposed algorthm are stated n pseudo code that shown n Fg. 3 and wll descrbe n the followng [11]-[13]: A. Generaton of Antbody Populaton The real codng representaton s adopted for the generaton of antbody populaton because t s accurate, effcent and closest to the real desgn space. Each antbody n the real codng s encoded as a vector of floatng pont numbers represents the postons of the sensor nodes. Table I shows the antbody representaton of N sensor nodes locatons (x, y). In the proposed algorthm, the populaton conssts of p s antbodes and the length of each one s 2N genes. Frst N genes represent the x postons and the second N genes represent the y postons TABLE I. Antbodes representaton of N sensor nodes locatons clonal prolferaton. Each gene n a sngle antbody, dependng on the hypermutaton rate p h, executes the hypermutaton of convex combnaton. The hypermutaton rate p h has an extremely hgh rate than the normal mutaton rate to ncrease the antbodes dversty. For a gven sensors postons antbody S ( L1, L2,..., L L,..., L2 N ), f the gene L s determned to execute the hypermutaton and another gene L s randomly selected to jon n, the resultng offsprng antbody becomes S ( L1, L2,..., L L,..., L2 N), where the new gene L s L 1 L L, and [0, 1 ] s a random value. Start BS requests the IDs, and ntal postons of all sensor nodes n the sensor feld Generate ntal populaton wth sze p s and defne p r, p c, p m, p h and Maxgen then set gen=0 Compute the objectve functon F(S) for all antbodes usng Equaton (10) Locatons x- locatons y- locatons L 1 L 2 L N L N+1 L N+2 L 2N Increase number of generatons by one (gen= gen+1) Postons Antbody (S) x 1 x 2 x N y 1 y 2 y N B. Objectve Functon Evaluaton The man purpose of the proposed algorthm s fndng the locatons of the MWSN after ntal random deployment to maxmze the covered area and mnmze the movng energy consumpton based on mnmzaton of the uncovered area rato f 1 (S) and movng dstances of all sensor nodes f 2 (S) as follows: ( ) = ( ) + (1 ) ( ) (10) The value of w (0 w 1) s applcaton-dependent. It ndcates whch factor s more mportant to be consdered. C. Selecton The roulette wheel selecton [12] s employed n mmune based algorthms for antbodes reproducton. Its basc dea s to determne the selecton probablty for each sensor s poston antbody n proporton to the ftness value (1/F(S)). The postons antbodes wth hgher ftness values are more lely to be selected as the parents antbodes that generate offsprngs n the next steps. D. Replcaton Replcaton s used to select better (p r p s ) sensors postons antbodes based on ts objectve functon F(S), where p r s the replcaton rate. Better antbodes are those have mnmum objectve functon values. E. Clonal Prolferaton wthn Hypermutaton On the bass of the bologcal mmune prncples, some postons antbodes dependng on the clonal selecton rate p c are chosen from the antbodes populaton pool to jon the No Apply roulette wheel selecton mechansm on all poston antbodes Apply clonal prolferaton wth hypermutaton Mutate the hypermutated and replcated antbodes Compute the F(S) for all new offsprngs usng Equaton (10) Select the better p s antbodes as new populaton of the next generaton Meet stoppng crteron? Yes BS sends the calculated postons to sensor nodes End Fg.2.Immune node deployment algorthm Replcate the selected antbodes 3
F. Mutaton Operaton The basc concept of mutaton operaton s also derved from convex set theory. Two genes n a sngle sensor poston antbody are randomly chosen to execute the mutaton combnaton. For a gven S ( L1, L2,..., L L,..., L 2N ), f the elements L and L are randomly selected for mutaton dependng on a mutaton rate p m, the resultng offsprng s S ( L1, L2,..., L L,..., L2N ). The two new genes L and L are L 1 L L L L 1 L respectvely, where β s selected randomly n [0, 1] range. and G. Selecton of the Best Solutons Better antbodes are always survved. In ths step, the ntal populaton pool (parents) and offsprngs antbodes that generated n prevous step are sorted n ascendng order based on the objectve functon values. Then the frst p s postons antbodes wth mnmum objectve functon values are selected to form the populaton pool for next generaton. H. Stoppng Crteron The stoppng crteron acheves when the objectve functon doesn t change for certan number of generatons or when the number of generatons exceeds the specfed maxmum generatons (Maxgen). Request (ID and ntal poston of nodes); % short message Set (N, Rs, feld sze); % Networ Intalzaton Set (p s, pr, pc, ph, pm, Maxgen) ; % IA parameters gen=0; % Intalzaton of generatons counter Chrom=Intal_pop(); % Construct the ntal populaton pool Whle (stoppng crteron false) gen=gen+1; % Increment the number of generatons Evaluuate (Chrom); % Objectve functon evaluaton % Roulette wheel selecton Chrom_sel=RWS_Selecton(Chrom); % Selecton of better antbodes usng Replcaton Chrom_rep=replcaton(Chrom_sel); Chrom_clon=Clonng(Chrom_rep); % Clonal operaton % Hypermutaton operaton Chrom_hyper=Hypermutaton(Chrom_clon); Chrom_tot=[ Chrom_rep, Chrom_hyper]; Chrom_chld=Mutaton(Chrom_tot); % Mutaton Operaton Evaluuate (Chrom_chld); % Objectve functon evaluaton % Selecton of better antbodes for next generaton Chrom=Better_selecton(Chrom, Chrom_chld); End Send (the calculated postons to the sensor nodes); Fg.3. Pseudo code of the proposed algorthm V. SIMULATION RESULTS In order to verfy the valdty of the proposed node deployment algorthm, the smulaton s mplemented usng Matlab to evaluate the proposed algorthm. Two experments are consdered here to compare the proposed node deployment algorthm wth the prevous algorthms that are descrbed n [7]- [8]. To elmnate the expermental error caused by randomness, each experment was run for 20 tmes and the average of results s calculated. A. Experment 1 To compare the proposed algorthm wth WSNPSO [7], the same number of sensor nodes and feld sze are adopted. Namely, the number of sensor nodes s vared from 20 to 60 wthn the same sensor feld szed 50x50 m 2 and wth fxed sensor coverage radus Rs=5 m as gven n Table II. The IA parameters adopted are set as p s =40, p r =0.9, p m =0.01, p c =0.1, p h =0.3, the weghtng factor w s 0.9 and the maxmum number of generaton (Maxgen) s 200. Fg.4 shows the ntal coverage, the deal coverage (( ) ) and the covered area (R Cov ) obtaned usng the proposed algorthm and WSNPSO [7]. From ths fgure, t s notced that the proposed algorthm mproves the coverage area for 5 tests by movng small number of sensor nodes short dstances as shown n Fg.5 ths reduces the movng consumpton energy. Ths fgure shows the maxmum moved dstance (d max ) and average moved dstance of all sensor nodes n each test obtaned usng the proposed algorthm compared to the maxmum moved dstances obtaned usng WSNPSO algorthm.. Fg.6 shows the coverage area for Test 1, Test 3 and Test 5 before and after deployment. The ponts mared by s are the ntal postons of the sensors whle those mared by s are ther fnal postons. Arrows are drawn to represent the sensors movement from ther ntal postons to ther fnal postons. From the results, t s notced that the proposed algorthm acheves the objectves of best deployment because t maxmzes the coverage area and also prolongs the operatonal lfe of the ndvdual nodes by reducng the movng consumpton energy. In Test 1, the proposed algorthm utlzes 20 sensors to cover 62.1727% of the sensor feld wth maxmum moved dstance 7.7260m and average moved dstance 2.9761 m but the WSNPSO covers 59.9014% of sensor feld wth maxmum moved dstance of 16.5961 m. Snce the number of sensor nodes ncreases to 60, the proposed algorthm ncreases the covered area to 99.4747% wth maxmum moved dstance (d max ) equals 8.8757 m and average moved dstance 1.9185 m but the WSNPSO covers 96.9578 wth d max equals 13.1925 m. Moreover, t s observed that the average dstance decreases wth hgher sensors densty. Ths s because a denser networ has more number of sensors, thus the sensors do not need to move too far to mprove the coverage as opposed to a sparse networ. TABLE II Networ specfcaton of experment 1 Test No. Sensor feld Rs No. of sensor nodes (N) Test 1 50x50 5 20 Test 2 50x50 5 30 Test 3 50x50 5 40 Test 4 50x50 5 50 Test 5 50x50 5 60 4
Fg.4. Comparson of coverage area for fve tests Fg.5. Moved dstance for fve tests (a) The ntal coverage networ before redeployment for tests 1, 3 and 5 respectvely (from left to rght) (b) The coverage networ after redeployment for tests 1, 3 and 5 respectvely (from left to rght) B. Experment 2 In ths experment, the performance of the proposed algorthm s compared wth the algorthm descrbed n [8]. For ths purpose the smulaton consders 23 randomly dstrbuted moble sensor nodes wthn 50x50 m 2 sensor feld and the sensor coverage radus Rs=7 m. The IA parameters adopted are set as p s =40, p r =0.9, p m =0.01, p c =0.1, p h =0.3, the weghtng factor w s 0.9 and Maxgen=200. The ntal and fnal sensor nodes postons are represented by s and s respectvely as Fg.6. Networ coverage for tests 1, 3 and 5 shown Fg. 4. Fgures 5 and 6 llustrate the covered area rato (R Cov ) and the redundant covered area versus number of generaton respectvely for the two algorthms. From these fgures, t can be observed that the proposed algorthm outperforms the algorthm n [8] n term of the covered area and the redundant covered area. After 10 generatons from startng, the coverage area reaches to 88.98 % usng the proposed algorthm and reaches to 80.1% usng the algorthm descrbed n [8]. On the other hand the redundant covered area reaches to 5
954 m 2 and 1103 m 2 respectvely for the proposed algorthm and the algorthm n [8]. The proposed algorthm ncreases the covered area gradually to 97.9% after 100 generatons and decreases the redundant covered area rapdly than the algorthm n [8] to 679.8 m 2, but the covered area and the redundant covered area usng the algorthm n [8] reach to 96.66% and 729.5 m 2 respectvely. After 200 generaton the proposed algorthm mproved the covered area by 1.94% and the redundant covered area by 13.31% than the algorthm n [8]. Furthermore, the proposed algorthm eeps the movng consumpton energy mnmum. (a) Before deployment (b) After deployment Fg.4.The coverage networ (experment 2) Fg.7.The covered area rato (experment 2) VI. CONCLUSION In ths paper, a new mmune node deployment algorthm for moble wreless sensor networ has been presented to mprove the networ coverage. The mmune algorthm s used to redeploy the moble sensor nodes after the ntal confguraton based on maxmzaton of the coverage area and mnmzaton of the movng dsspated energy. Smulaton results showed that the proposed algorthm outperforms the other algorthms n term of the coverage networ and the redundant covered area. Moreover, t mnmzes the movng dsspated energy by mnmzng the root mean of the sum squared moved dstances of the moble sensors. Fnally, we concluded that the moblty of sensor nodes has major advantages on statc ones n enhancng the networ coverage and reducng the redundant covered area. REFERENCES [1] Y. Qu and S.V. Georgaopoulos, Relocaton of Wreless Sensor Networ Nodes Usng a Genetc Algorthm,Proceedngs of 12th Annual IEEE Wrelessand Mcrowave Technology Conference (WAMICON), Clearwater Beach, 18-19 Aprl 2011, pp. 1-5. [2] B. Wang, Coverage Problems n Sensor Networs: A Survey, ACM Computng Surveys, Vol. 43, No. 4, 2011. [3] O. Banmelhem, M. Mowaf and W. Aljoby, "Genetc Algorthm Based Node Deployment n Hybrd Wreless Sensor Networs," Communcatons and Networ, vol. 5, no. 4, pp. 273-279, 2013. [4] W. Xaolng, S. Le, W. Jn, J. Cho1and S.Lee, Energy Effcent Deployment of Moble Sensor Networs by PSO, Advanced Web and Networ Technologes and Applcatons,Lecture Notes n Computer Scence, vol. 3842, pp 373-382,2006. [5] N.Heo and P. K. Varshney, Energy-Effcent Deployment of Intellgent Moble Sensor Networs,IEEE Transactons on Systems, Man, and Cybernetcs Part A: Systems and Humans, vol. 35, no. 1, pp. 78-92,January 2005. [6] A Norouz, F. S. Babamr, and A. H.Zam, An Interactve Genetc Algorthm for Moble Sensor Networs, Studes n Informatcs and Control, ISSN 1220-1766, vol. 22, no. 2, pp. 213-218, 2013. [7] N Ab. Azz, A. W. Mohemmed, M. Y. Alas, K. Ab. Azz and S. Syahal, Coverage Maxmzaton and Energy Conservaton for Moble Wreless Sensor Networs: A Two Phase Partcle Swarm Optmzaton Algorthm, Internatonal Journal of Natural Computng Research (IJNCR), vol. 3, no. 2, pp., 2012. [8] L. Jn, J. Ja and D. Sun, Node Dstrbuton Optmzaton n Moble Sensor Networ Based on MultObjectve Dfferental Evoluton Algorthm, Fourth Internatonal Conference on Genetc and Evolutonary Computng, Shenzhen, Chna, 13-15 December 2010. [9] D. Deepanandhn, and T. Amudha, Solvng Job Shop Schedulng Problems Wth Consultant Guded Search Metaheurstcs, Internatonal Journal of Software and Web Scences, vol. 3, no. 1, pp. 1-6, Feb. 2013. [10] Deyng L, Ha Lu, Sensor Coverage n Wreless Sensor Networs, Wreless Networs: Research, Technology and Applcatons, Ja Feng (ed.), Nova Scence Publshers, pp. 3-31, 2009. [11] X. Lu, Y. Dng and K.Hao, Immune Clonal Selecton Algorthm for Target Coverage of Wreless Sensor Networs, Int. J. Modellng, Identfcaton and Control, vol. 12, no. 1, pp. 119-124,January 2011. [12] M. Abo-Zahhad, S. M. Ahmed, N. Sabor and A. F. Al-Ajloun, "Desgn of Two-Dmensonal Recursve Dgtal Flters wth Specfed Magntude and Group Delay Characterstcs usng Taguch-based Immune Algorthm", Int. J. of Sgnal and Imagng Systems Engneerng, vol. 3, no. 3, pp. 222-235, January 2010. [13] M. Abo-Zahhad, S. M. Ahmed, N. Sabor and A. F. Al-Ajloun, " The Convergence Speed of Sngle-And Mult-Objectve Immune Algorthm Based Optmzaton Problems", SgnalProcessng: An InternatonalJournal, vol. 4, no. 5, pp. 247-266, 2010. Fg.8.The redundant covered area (experment 2) 6